From Silence to Signal: Extracting E-Commerce Feedback With AI-Driven Personalized Surveys

The Problem: Low Actionable Feedback Rates Here's a common pattern in e-commerce feedback (and these are realistic for most platforms): Launch a product → 10,000 customers buy → 10–20% leave reviews → 20–30% of those contain actionable insights. Launch a product → 10,000 customers buy → 10–20% leave reviews → 20–30% of those contain actionable insights. Familiar products like electronics or apparel tend to sit toward the higher end of review volume, while obscure or untrusted products end up with fewer, more uneven reviews. Product decisions often rely on feedback from roughly 2–6% of the shopper base. The silent 94–98% remain largely unknown. Writing reviews requires effort, which naturally filters out customers without strong opinions. Product decisions often rely on feedback from roughly 2–6% of the shopper base. Why traditional feedback struggles? Why traditional feedback struggles? What users experience: What users experience: High cognitive load: Writing a coherent review takes 2–3 minutes. Most customers won't invest this time unless they feel strongly committed. Generic forms: The same feedback prompts—value for money, overall satisfaction, star ratings—are used for everything, whether it’s a smartwatch or a bath mat. Uncertain impact: Submit feedback → Hear nothing → Product may not improve → Reduced motivation to provide future feedback. High cognitive load: Writing a coherent review takes 2–3 minutes. Most customers won't invest this time unless they feel strongly committed. High cognitive load: Generic forms: The same feedback prompts—value for money, overall satisfaction, star ratings—are used for everything, whether it’s a smartwatch or a bath mat. Generic forms: value for money, overall satisfaction, star ratings Uncertain impact: Submit feedback → Hear nothing → Product may not improve → Reduced motivation to provide future feedback. Uncertain impact: What businesses see: What businesses see: Selection bias: The ecstatic (5-star) and frustrated (1-star) customers are most likely to write reviews. The nuanced middle ("Good, but the battery could be better") usually stay silent. Low-signal reviews: Many reviews are vague—“Great product!” or “Terrible quality”—with no details that teams can actually act on. Star ratings without context: A 3-star review could mean "Mediocre product" or "Great product with one fixable flaw." Understanding the difference requires manually writing the reviews. Delayed pattern detection: By the time a pattern becomes obvious in reviews, thousands—or tens of thousands—more units may already be in customers' hands with the same issue. Limited follow-up: "Battery life is poor" raises questions it can't answer. How long does it last? What were you doing? What did you expect? Written feedback rarely allows deeper probing unless the user chooses to elaborate. Selection bias: The ecstatic (5-star) and frustrated (1-star) customers are most likely to write reviews. The nuanced middle ("Good, but the battery could be better") usually stay silent. Selection bias: Low-signal reviews: Many reviews are vague—“Great product!” or “Terrible quality”—with no details that teams can actually act on. Low-signal reviews: Star ratings without context: A 3-star review could mean "Mediocre product" or "Great product with one fixable flaw." Understanding the difference requires manually writing the reviews. Star ratings without context: Delayed pattern detection: By the time a pattern becomes obvious in reviews, thousands—or tens of thousands—more units may already be in customers' hands with the same issue. Delayed pattern detection: Limited follow-up: "Battery life is poor" raises questions it can't answer. How long does it last? What were you doing? What did you expect? Written feedback rarely allows deeper probing unless the user chooses to elaborate. Limited follow-up: How long does it last? What were you doing? What did you expect? The Solution: Intelligent MCQ Interviews What if we could extract high-signal feedback effortlessly—driving product improvement, building platform trust, and making it easy for users to shape better products? Instead of asking users to write reviews, we interview them with hyper-personalized MCQs. Not generic forms. Not hardcoded decision trees. Adaptive, context-aware questions that are shaped by: What if we could extract high-signal feedback effortlessly—driving product improvement, building platform trust, and making it easy for users to shape better products? write interview Adaptive, context-aware questions Product intelligence: What are the known pain points for this product? (Analyzed from existing reviews and description) User intelligence: What's their purchase history? How do they typically review? (Behavioral profiling) Peer intelligence: What did similar buyers complain about? (Vector similarity across cohorts) Product intelligence: What are the known pain points for this product? (Analyzed from existing reviews and description) Product intelligence: this product User intelligence: What's their purchase history? How do they typically review? (Behavioral profiling) User intelligence: Peer intelligence: What did similar buyers complain about? (Vector similarity across cohorts) Peer intelligence: The core insight: MCQs collapse cognitive load while preserving signal. One-click responses can generate as much or more depth than written reviews—if the questions are smart enough. The core insight: MCQs collapse cognitive load while preserving signal. if The potential value for: The potential value for: Businesses: More diverse feedback through guided interviews, faster pattern detection for product iteration Platforms: Increased engagement through lower-friction feedback collection, richer review data Users: Interactive MCQs replace lengthy text writing, direct input into product improvements Businesses: More diverse feedback through guided interviews, faster pattern detection for product iteration Businesses: Platforms: Increased engagement through lower-friction feedback collection, richer review data Platforms: Users: Interactive MCQs replace lengthy text writing, direct input into product improvements Users: Introducing Survey Sensei—a multi-agent system that implements this approach with four specialized agents: Survey Sensei ProductContextAgent → Analyzes reviews/metadata to understand product strengths and weaknesses CustomerContextAgent → Builds behavioral profile from purchase history and review patterns SurveyAgent (adaptive MCQ engine) → Generates tailored questions, reshapes follow-ups based on each answer ReviewGenAgent → Synthesizes MCQ selections into natural language reviews in the user's voice ProductContextAgent → Analyzes reviews/metadata to understand product strengths and weaknesses ProductContextAgent CustomerContextAgent → Builds behavioral profile from purchase history and review patterns CustomerContextAgent SurveyAgent (adaptive MCQ engine) → Generates tailored questions, reshapes follow-ups based on each answer SurveyAgent (adaptive MCQ engine) ReviewGenAgent → Synthesizes MCQ selections into natural language reviews in the user's voice ReviewGenAgent The full workflow (end-to-end example): The full workflow (end-to-end example): Step 1 (Product Intelligence): Analyze 213 reviews of this laptop → Battery: 67% complain "dies mid-afternoon" → Keyboard: 82% praise "excellent typing experience" → Performance: 45% mention "handles multitasking well" Step 2 (User Intelligence): Pull purchase history for this customer → Bought 3 laptops in past 2 years (power user pattern) → Reviews 85% of purchases, critical but fair (3.6★ average) → Detail-oriented: past reviews averaged 120 words Step 3 (Adaptive Question 1): "You've purchased 3 laptops recently. What drove this upgrade?" ○ Better performance ● Longer battery life ← USER SELECTED ○ Lighter/more portable ○ Other: [text] Step 4 (Adaptive Follow-up): "How long does the battery last on a typical workday?" ○ All day (12+ hours) ● 4-8 hours ← USER SELECTED ○ Less than 4 hours Step 5 (Probing Deeper): "Does this meet your battery expectations?" ○ Exceeds expectations ● Falls slightly short ← USER SELECTED ○ Major disappointment [Agent continues for 10-12 total questions, probing keyboard quality, performance, portability based on this user's priorities...] Step 6 (Review Synthesis): Convert MCQ selections → natural language "Upgraded hoping for better battery. Lasts 4-8 hours—falls short of all-day claims, but manageable for office work. Keyboard is outstanding for typing. Performance handles multitasking well." Step 1 (Product Intelligence): Analyze 213 reviews of this laptop → Battery: 67% complain "dies mid-afternoon" → Keyboard: 82% praise "excellent typing experience" → Performance: 45% mention "handles multitasking well" Step 2 (User Intelligence): Pull purchase history for this customer → Bought 3 laptops in past 2 years (power user pattern) → Reviews 85% of purchases, critical but fair (3.6★ average) → Detail-oriented: past reviews averaged 120 words Step 3 (Adaptive Question 1): "You've purchased 3 laptops recently. What drove this upgrade?" ○ Better performance ● Longer battery life ← USER SELECTED ○ Lighter/more portable ○ Other: [text] Step 4 (Adaptive Follow-up): "How long does the battery last on a typical workday?" ○ All day (12+ hours) ● 4-8 hours ← USER SELECTED ○ Less than 4 hours Step 5 (Probing Deeper): "Does this meet your battery expectations?" ○ Exceeds expectations ● Falls slightly short ← USER SELECTED ○ Major disappointment [Agent continues for 10-12 total questions, probing keyboard quality, performance, portability based on this user's priorities...] Step 6 (Review Synthesis): Convert MCQ selections → natural language "Upgraded hoping for better battery. Lasts 4-8 hours—falls short of all-day claims, but manageable for office work. Keyboard is outstanding for typing. Performance handles multitasking well." Contrast with generic tools: Contrast with generic tools: Generic survey: "Rate this product 1-5. Any comments?" (No context and high inertia) Survey Sensei: Adaptive interview that knows you're a repeat buyer with battery concerns and generates 10-12 contextual MCQs to extract nuanced feedback in a minute. Generic survey: "Rate this product 1-5. Any comments?" (No context and high inertia) Generic survey: Survey Sensei: Adaptive interview that knows you're a repeat buyer with battery concerns and generates 10-12 contextual MCQs to extract nuanced feedback in a minute. Survey Sensei: Why Now? What Changed to Make This Possible The convergence of cheaper, more intelligent models with rapidly declining token costs has made AI-powered personalization economically viable at scale. 2020 (GPT-3 era): 2020 (GPT-3 era): Language models: Variable outputs with heavy reliance on prompt engineering Embeddings: Early-stage and largely experimental, with limited production infrastructure Stateful agents: Custom-built state machines for each workflow Economics: Relatively higher token costs made multi-step, interactive workflows difficult to justify at scale Language models: Variable outputs with heavy reliance on prompt engineering Embeddings: Early-stage and largely experimental, with limited production infrastructure Stateful agents: Custom-built state machines for each workflow Economics: Relatively higher token costs made multi-step, interactive workflows difficult to justify at scale 2025 (GPT-4 / GPT-5 era and evolving): 2025 (GPT-4 / GPT-5 era and evolving): Language models: Reliable reasoning and structured outputs (this prototype uses GPT-4o-mini but advanced models can also be leveraged) Embeddings & retrieval: Production-ready vector databases with mature tooling Agent frameworks: First-class support for stateful, multi-step workflows Economics: Token costs low enough for personalized, interactive flows at scale Language models: Reliable reasoning and structured outputs (this prototype uses GPT-4o-mini but advanced models can also be leveraged) Embeddings & retrieval: Production-ready vector databases with mature tooling Agent frameworks: First-class support for stateful, multi-step workflows Economics: Token costs low enough for personalized, interactive flows at scale What This Enables 1. Per-user personalization: 1. Per-user personalization: Questions adapt to purchase history and behavioral patterns Depth and complexity match user engagement level Topic selection reflects individual product concerns Questions adapt to purchase history and behavioral patterns Depth and complexity match user engagement level Topic selection reflects individual product concerns 2. Adaptive vs. static workflows: 2. Adaptive vs. static workflows: Traditional: Fixed question sequences regardless of responses AI-powered: Follow-up questions probe deeper based on answers Traditional: Fixed question sequences regardless of responses AI-powered: Follow-up questions probe deeper based on answers 3. Natural language synthesis: 3. Natural language synthesis: Traditional templates: Generic phrasing, obvious patterns AI synthesis: Contextual details, varied expression Traditional templates: Generic phrasing, obvious patterns AI synthesis: Contextual details, varied expression 4. Economic accessibility: 4. Economic accessibility: Earlier LLM costs limited surveys to high-value scenarios Current pricing makes this solution viable Lower costs enable experimentation and iteration Earlier LLM costs limited surveys to high-value scenarios Current pricing makes this solution viable Lower costs enable experimentation and iteration Two-Part Architecture Before diving into the details, it's critical to understand how the project is structured. The diagram below shows the complete system architecture—from the UI layer through the orchestrator to the multi-agent framework, along with data pipelines and database schema: Before diving into the details, it's critical to understand how the project is structured: Part 1: Simulation Infrastructure (Testing Layer) Purpose: Development scaffolding—to test the core system without production data. Purpose: MockDataOrchestrator creates a semi-realistic sample of an e-commerce ecosystem. For example: MockDataOrchestrator Products: RapidAPI fetch (real Amazon data) + 5 similar products and 3 diverse products (mocked) Users: Main user + N mock personas Reviews: RapidAPI reviews (real) + LLM-generated reviews Transactions: 40-60% have reviews (matches reality) Embeddings: Generated in batch, parallelized for efficiency Products: RapidAPI fetch (real Amazon data) + 5 similar products and 3 diverse products (mocked) Users: Main user + N mock personas Reviews: RapidAPI reviews (real) + LLM-generated reviews Transactions: 40-60% have reviews (matches reality) Embeddings: Generated in batch, parallelized for efficiency In production: Skip this entirely and integrate with real e-commerce databases and pipelines. In production: Part 2: Agentic Survey Framework (Core USP) Purpose: The actual product—adaptive survey generation + authentic review synthesis. Purpose: This is the heart of Survey Sensei. The 4-agent system decomposes into specialized agents, each with focused responsibility: Agent 1: ProductContextAgent Agent 1: ProductContextAgent Builds a mental model of the product before generating questions. Three-path adaptive logic: Three-path adaptive logic: Direct Reviews Path (Confidence: 70-95%) Condition: Product already has reviews in database Ranking heuristic: Recency (50% weight, exponential decay with 180-day half-life) + Quality (40%, review length) + Diversity (10%, bonus for 3-4 star reviews) Confidence formula: 0.70 + (num_reviews / 100), capped at 0.95 Extracts: Key features, pain points, use cases, pros/cons, sentiment patterns Similar Products Path (Confidence: 55-80%) Condition: No reviews for this product, but vector-similar products exist Process: Cosine similarity search via pgvector (threshold: 0.7) Ranking heuristic: Similarity (40%) + Recency (35%) + Quality (20%) + Diversity (5%) Extracts: Inferred experience from analogous products Generic/Description Only Path (Confidence: 40-50%) Condition: New product, zero reviews anywhere Process: Parse title, description, category metadata Extracts: Educated guesses (e.g., "Wireless device → ask about battery") Direct Reviews Path (Confidence: 70-95%) Condition: Product already has reviews in database Ranking heuristic: Recency (50% weight, exponential decay with 180-day half-life) + Quality (40%, review length) + Diversity (10%, bonus for 3-4 star reviews) Confidence formula: 0.70 + (num_reviews / 100), capped at 0.95 Extracts: Key features, pain points, use cases, pros/cons, sentiment patterns Direct Reviews Path Condition: Product already has reviews in database Ranking heuristic: Recency (50% weight, exponential decay with 180-day half-life) + Quality (40%, review length) + Diversity (10%, bonus for 3-4 star reviews) Confidence formula: 0.70 + (num_reviews / 100), capped at 0.95 Extracts: Key features, pain points, use cases, pros/cons, sentiment patterns Condition: Product already has reviews in database Ranking heuristic: Recency (50% weight, exponential decay with 180-day half-life) + Quality (40%, review length) + Diversity (10%, bonus for 3-4 star reviews) Confidence formula: 0.70 + (num_reviews / 100), capped at 0.95 Extracts: Key features, pain points, use cases, pros/cons, sentiment patterns Similar Products Path (Confidence: 55-80%) Condition: No reviews for this product, but vector-similar products exist Process: Cosine similarity search via pgvector (threshold: 0.7) Ranking heuristic: Similarity (40%) + Recency (35%) + Quality (20%) + Diversity (5%) Extracts: Inferred experience from analogous products Similar Products Path Condition: No reviews for this product, but vector-similar products exist Process: Cosine similarity search via pgvector (threshold: 0.7) Ranking heuristic: Similarity (40%) + Recency (35%) + Quality (20%) + Diversity (5%) Extracts: Inferred experience from analogous products Condition: No reviews for this product, but vector-similar products exist this Process: Cosine similarity search via pgvector (threshold: 0.7) Ranking heuristic: Similarity (40%) + Recency (35%) + Quality (20%) + Diversity (5%) Extracts: Inferred experience from analogous products Generic/Description Only Path (Confidence: 40-50%) Condition: New product, zero reviews anywhere Process: Parse title, description, category metadata Extracts: Educated guesses (e.g., "Wireless device → ask about battery") Generic/Description Only Path Condition: New product, zero reviews anywhere Process: Parse title, description, category metadata Extracts: Educated guesses (e.g., "Wireless device → ask about battery") Condition: New product, zero reviews anywhere Process: Parse title, description, category metadata Extracts: Educated guesses (e.g., "Wireless device → ask about battery") Output schema: Output schema: class ProductContext: key_features: List[str] major_concerns: List[str] pros: List[str] cons: List[str] common_use_cases: List[str] context_type: str confidence_score: float class ProductContext: key_features: List[str] major_concerns: List[str] pros: List[str] cons: List[str] common_use_cases: List[str] context_type: str confidence_score: float Agent 2: CustomerContextAgent Agent 2: CustomerContextAgent Builds a behavioral profile to personalize question depth and tone. Three-path adaptive logic: Three-path adaptive logic: Exact Interaction Path (Confidence: 85-95%) User bought THIS exact product before Ground truth on what they thought Similar Products Path (Confidence: 55-80%) Ranking heuristic: Similarity (45%) + Recency (30%) + Engagement (25%) Infer preferences from purchase patterns (e.g., "Bought 3 noise-canceling headphones → cares about ANC quality") Demographics Path (Confidence: 35-45%) Brand new user, zero purchase history Generic baseline persona Exact Interaction Path (Confidence: 85-95%) User bought THIS exact product before Ground truth on what they thought Exact Interaction Path User bought THIS exact product before Ground truth on what they thought User bought THIS exact product before Ground truth on what they thought Similar Products Path (Confidence: 55-80%) Ranking heuristic: Similarity (45%) + Recency (30%) + Engagement (25%) Infer preferences from purchase patterns (e.g., "Bought 3 noise-canceling headphones → cares about ANC quality") Similar Products Path Ranking heuristic: Similarity (45%) + Recency (30%) + Engagement (25%) Infer preferences from purchase patterns (e.g., "Bought 3 noise-canceling headphones → cares about ANC quality") Ranking heuristic: Similarity (45%) + Recency (30%) + Engagement (25%) Ranking heuristic: Infer preferences from purchase patterns (e.g., "Bought 3 noise-canceling headphones → cares about ANC quality") Demographics Path (Confidence: 35-45%) Brand new user, zero purchase history Generic baseline persona Demographics Path Brand new user, zero purchase history Generic baseline persona Brand new user, zero purchase history Generic baseline persona Output schema: Output schema: class CustomerContext: purchase_patterns: List[str] review_behavior: List[str] product_preferences: List[str] primary_concerns: List[str] expectations: List[str] pain_points: List[str] engagement_level: str # highly_engaged | moderately_engaged | passive_buyer | new_user sentiment_tendency: str # positive | critical | balanced | polarized | neutral review_engagement_rate: float confidence_score: float class CustomerContext: purchase_patterns: List[str] review_behavior: List[str] product_preferences: List[str] primary_concerns: List[str] expectations: List[str] pain_points: List[str] engagement_level: str # highly_engaged | moderately_engaged | passive_buyer | new_user sentiment_tendency: str # positive | critical | balanced | polarized | neutral review_engagement_rate: float confidence_score: float Personalization: Personalization: Critical + highly engaged → Deep technical MCQs Passive buyer → Simple MCQs Critical + highly engaged → Deep technical MCQs Passive buyer → Simple MCQs Agent 3: SurveyAgent (Stateful) Agent 3: SurveyAgent (Stateful) Conducts adaptive surveys where questions evolve based on answers. Uses LangGraph StateGraph for conversation state. Questions are generated in mini-batches to optimize API costs while maintaining adaptive flexibility. Performance optimization: Survey state is kept in memory during the session and written to the database at two points: LangGraph StateGraph Performance optimization: Survey start: Initial contexts frozen to product_context and customer_context JSONB columns Survey completion: Final Q&A written to questions_and_answers JSONB, complete state to session_context JSONB Survey start: Initial contexts frozen to product_context and customer_context JSONB columns Survey start: product_context customer_context Survey completion: Final Q&A written to questions_and_answers JSONB, complete state to session_context JSONB Survey completion: questions_and_answers session_context All intermediate answers are logged asynchronously to survey_details table for analytics (fire-and-forget, non-blocking). Here’s how the survey process is structured: survey_details Survey completion rules: Survey completion rules: initial_questions_count: 3 # Start with 3 baseline MCQs min_answered_questions: 10 # User must answer ≥10 max_answered_questions: 15 # Hard stop at 15 max_survey_questions: 20 # Total questions asked max_consecutive_skips: 3 # 3 consecutive skips → must answer to continue initial_questions_count: 3 # Start with 3 baseline MCQs min_answered_questions: 10 # User must answer ≥10 max_answered_questions: 15 # Hard stop at 15 max_survey_questions: 20 # Total questions asked max_consecutive_skips: 3 # 3 consecutive skips → must answer to continue Adaptive questioning example: Adaptive questioning example: Question 5: "How long does the battery last on a typical workday?" ● Less than 4 hours ← USER SELECTED [Agent's internal state update: - Battery performance: Below average - Action: Generate follow-up to quantify impact] Follow-up Question 6: "When does the battery typically die?" ● Mid-afternoon (2-4pm) ← USER SELECTED [Agent's internal state update: - Specific pain point: Dies at 2-4pm (work hours) - Severity: High (impacts productivity) - Action: Probe importance for review weighting] Follow-up Question 7: "How important is longer battery life to you?" ● Very important - major inconvenience (USER SELECTED) Question 5: "How long does the battery last on a typical workday?" ● Less than 4 hours ← USER SELECTED [Agent's internal state update: - Battery performance: Below average - Action: Generate follow-up to quantify impact] Follow-up Question 6: "When does the battery typically die?" ● Mid-afternoon (2-4pm) ← USER SELECTED [Agent's internal state update: - Specific pain point: Dies at 2-4pm (work hours) - Severity: High (impacts productivity) - Action: Probe importance for review weighting] Follow-up Question 7: "How important is longer battery life to you?" ● Very important - major inconvenience (USER SELECTED) Why adaptive matters: Without adaptive AI, feedback systems rely on rigid decision trees like ”Battery life: Excellent | Good | Fair | Poor”. That tells you what someone thinks, but not why. What’s missing is the detail that actually matters: “It dies at 2pm during work hours, which makes it unusable for me.” Adaptive AI doesn’t just branch. It regenerates the next question based on what the user has already said, pulling context forward instead of flattening it. Why adaptive matters: Battery life: Excellent | Good | Fair | Poor” what why “It dies at 2pm during work hours, which makes it unusable for me.” Agent 4: ReviewGenAgent Agent 4: ReviewGenAgent Convert MCQ selections into natural language authentic reviews matching user's writing style. Three-stage synthesis: Three-stage synthesis: Sentiment Classification → Analyze MCQ answers → Classify as good | okay | bad Voice Matching → Fetch historical reviews → Extract tone, vocabulary, sentence structure Generate 2–3 variations with different star ratings within the sentiment band: Sentiment Classification → Analyze MCQ answers → Classify as good | okay | bad Sentiment Classification good | okay | bad Voice Matching → Fetch historical reviews → Extract tone, vocabulary, sentence structure Voice Matching Generate 2–3 variations with different star ratings within the sentiment band: Generate 2–3 variations def _get_star_ratings(sentiment_band: str) -> List[int]: if sentiment_band == "good": return [5, 4] elif sentiment_band == "okay": return [4, 3, 2] else: # bad return [2, 1] def _get_star_ratings(sentiment_band: str) -> List[int]: if sentiment_band == "good": return [5, 4] elif sentiment_band == "okay": return [4, 3, 2] else: # bad return [2, 1] Example output (sentiment: "okay", user: concise + critical): Example output [4-star] "Solid build quality and excellent screen. Battery dies around 3pm—acceptable for office use where I have charging access. Keyboard is comfortable for long typing. Performance handles multitasking well. Worth it on sale." [3-star] "Mixed feelings. Build quality and screen are great, but battery is the main letdown—dies at 3pm despite 'all-day' claims. Keyboard is excellent. If battery isn't a dealbreaker, it's decent." [2-star] "Disappointed with battery life. Product page advertised all-day battery, but it dies by 3pm daily with moderate use. Screen and keyboard are good, but battery is a major problem for anyone working away from chargers." [4-star] "Solid build quality and excellent screen. Battery dies around 3pm—acceptable for office use where I have charging access. Keyboard is comfortable for long typing. Performance handles multitasking well. Worth it on sale." [3-star] "Mixed feelings. Build quality and screen are great, but battery is the main letdown—dies at 3pm despite 'all-day' claims. Keyboard is excellent. If battery isn't a dealbreaker, it's decent." [2-star] "Disappointed with battery life. Product page advertised all-day battery, but it dies by 3pm daily with moderate use. Screen and keyboard are good, but battery is a major problem for anyone working away from chargers." User picks framing, edits if needed, submits. Technical Implementation This section lists the tech stack used in this project and its main components. Backend: FastAPI (Python 3.11), LangChain + LangGraph, OpenAI GPT-4o-mini (Input: $0.15/1M, Output: $0.60/1M tokens), Pydantic, Supabase/PostgreSQL + pgvector Backend: Frontend: Next.js 14, TypeScript, Tailwind CSS, Supabase Client Frontend: AI/ML: OpenAI embeddings (1536-dim), batch generation (100 texts in 2-3s), IVFFlat indexes (2-3% recall loss for 100x speed) AI/ML: Database Schema -- 1. PRODUCTS: Catalog with semantic embeddings products ( item_id VARCHAR(20) PRIMARY KEY, title, brand, description, price, star_rating, num_ratings, review_count INTEGER, embeddings vector(1536), -- Semantic search is_mock BOOLEAN ) -- 2. USERS: Behavioral profiles users ( user_id UUID PRIMARY KEY, user_name, email_id, age, gender, base_location, embeddings vector(1536), total_purchases INTEGER, total_reviews INTEGER, review_engagement_rate DECIMAL(4,3), avg_review_rating DECIMAL(3,2), sentiment_tendency VARCHAR(20), engagement_level VARCHAR(30), is_main_user BOOLEAN ) -- 3. TRANSACTIONS: Purchase history transactions ( transaction_id UUID PRIMARY KEY, item_id → products, user_id → users, order_date, delivery_date, original_price, retail_price, transaction_status ) -- 4. REVIEWS: Multi-source feedback reviews ( review_id UUID PRIMARY KEY, item_id → products, user_id → users, transaction_id → transactions, review_title, review_text, review_stars, source VARCHAR(20), -- 'rapidapi' | 'agent_generated' | 'user_survey' embeddings vector(1536) ) -- 5. SURVEY_SESSIONS: Stateful survey orchestration survey_sessions ( session_id UUID PRIMARY KEY, user_id, item_id, transaction_id, product_context JSONB, -- Agent 1 output customer_context JSONB, -- Agent 2 output session_context JSONB, -- LangGraph state questions_and_answers JSONB, review_options JSONB, status VARCHAR(20) ) -- 6. SURVEY_DETAILS: Event log survey_details ( detail_id UUID PRIMARY KEY, session_id → survey_sessions, event_type VARCHAR(50), event_detail JSONB, created_at TIMESTAMP ) -- 1. PRODUCTS: Catalog with semantic embeddings products ( item_id VARCHAR(20) PRIMARY KEY, title, brand, description, price, star_rating, num_ratings, review_count INTEGER, embeddings vector(1536), -- Semantic search is_mock BOOLEAN ) -- 2. USERS: Behavioral profiles users ( user_id UUID PRIMARY KEY, user_name, email_id, age, gender, base_location, embeddings vector(1536), total_purchases INTEGER, total_reviews INTEGER, review_engagement_rate DECIMAL(4,3), avg_review_rating DECIMAL(3,2), sentiment_tendency VARCHAR(20), engagement_level VARCHAR(30), is_main_user BOOLEAN ) -- 3. TRANSACTIONS: Purchase history transactions ( transaction_id UUID PRIMARY KEY, item_id → products, user_id → users, order_date, delivery_date, original_price, retail_price, transaction_status ) -- 4. REVIEWS: Multi-source feedback reviews ( review_id UUID PRIMARY KEY, item_id → products, user_id → users, transaction_id → transactions, review_title, review_text, review_stars, source VARCHAR(20), -- 'rapidapi' | 'agent_generated' | 'user_survey' embeddings vector(1536) ) -- 5. SURVEY_SESSIONS: Stateful survey orchestration survey_sessions ( session_id UUID PRIMARY KEY, user_id, item_id, transaction_id, product_context JSONB, -- Agent 1 output customer_context JSONB, -- Agent 2 output session_context JSONB, -- LangGraph state questions_and_answers JSONB, review_options JSONB, status VARCHAR(20) ) -- 6. SURVEY_DETAILS: Event log survey_details ( detail_id UUID PRIMARY KEY, session_id → survey_sessions, event_type VARCHAR(50), event_detail JSONB, created_at TIMESTAMP ) Design decisions: Design decisions: JSONB for flexibility → Agent outputs evolve without migrations Vector indexes → IVFFlat gives 100x speed for 2-3% recall loss Source tracking → rapidapi (real) | agent_generated (mock) | user_survey (golden path) Event sourcing → survey_details logs every interaction for debugging JSONB for flexibility → Agent outputs evolve without migrations JSONB for flexibility Vector indexes → IVFFlat gives 100x speed for 2-3% recall loss Vector indexes Source tracking → rapidapi (real) | agent_generated (mock) | user_survey (golden path) Source tracking rapidapi agent_generated user_survey Event sourcing → survey_details logs every interaction for debugging Event sourcing survey_details Vector Similarity All text → 1536-dim embeddings via text-embedding-3-small. text-embedding-3-small Find similar products: Find similar products: SELECT item_id, title, 1 - (embeddings query_embedding) AS similarity FROM products WHERE 1 - (embeddings query_embedding) > 0.7 ORDER BY similarity DESC LIMIT 5; SELECT item_id, title, 1 - (embeddings query_embedding) AS similarity FROM products WHERE 1 - (embeddings query_embedding) > 0.7 ORDER BY similarity DESC LIMIT 5; Why vectors beat traditional categories: Why vectors beat traditional categories: Traditional hierarchies (Electronics → Audio → Headphones → Wireless) are rigid. Vector embeddings cluster products by intent and use case: intent and use case Category-based: "Wireless headphones" → Returns ALL wireless headphones Vector-based: "Premium noise-canceling headphones for travel" → Returns semantically similar products solving the same problem (Bose QC45, AirPods Max, Sennheiser Momentum 4) Category-based: "Wireless headphones" → Returns ALL wireless headphones Category-based: Vector-based: "Premium noise-canceling headphones for travel" → Returns semantically similar products solving the same problem (Bose QC45, AirPods Max, Sennheiser Momentum 4) Vector-based: Vector embeddings naturally cluster by intent rather than superficial attributes. "Noise-canceling Bluetooth headphones" is closer to "wireless earbuds with ANC" than to "studio monitor headphones"—even though all three are technically "headphones." intent Performance benchmarks: Performance benchmarks: Vector search (IVFFlat): 10,000 products in ~50ms Batch embeddings: 100 texts in 2-3 seconds Brute-force: 10,000 products in ~5 seconds (100× slower) Vector search (IVFFlat): 10,000 products in ~50ms Batch embeddings: 100 texts in 2-3 seconds Brute-force: 10,000 products in ~5 seconds (100× slower) API Design The API provides six endpoints that cover the end-to-end survey workflow: Error handling and edge cases: Error handling and edge cases: 1. Session expiration: 1. Session expiration: Survey sessions expire after 24 hours of inactivity Prevents abandoned surveys from cluttering the database User can't submit answers to expired sessions (returns HTTP 410 Gone) Survey sessions expire after 24 hours of inactivity Prevents abandoned surveys from cluttering the database User can't submit answers to expired sessions (returns HTTP 410 Gone) 2. Idempotency: 2. Idempotency: Answering the same question twice → Updates the answer (no duplicates) Submitting the same review twice → Ignored (review already posted) Prevents accidental double-submissions from network retries Answering the same question twice → Updates the answer (no duplicates) Submitting the same review twice → Ignored (review already posted) Prevents accidental double-submissions from network retries 3. Pydantic validation: 3. Pydantic validation: All API requests/responses validated with Pydantic schemas Fail fast: Invalid data rejected at API boundary (before hitting agents) Example: answer must be one of the provided options, not arbitrary text All API requests/responses validated with Pydantic schemas Fail fast: Invalid data rejected at API boundary (before hitting agents) Example: answer must be one of the provided options, not arbitrary text answer Running It Locally Source: github.com/arnavvj/survey-sensei Source: github.com/arnavvj/survey-sensei Prerequisites: Python 3.11+, Node.js 18+, Supabase project (free), OpenAI API key (~$5 credit) Prerequisites: Backend Setup 1. Clone the repo and create a python environment: git clone https://github.com/arnavvj/survey-sensei.git cd survey-sensei/backend conda env create -f environment.yml # Installs all deps (FastAPI, LangChain, etc.) conda activate survey-sensei git clone https://github.com/arnavvj/survey-sensei.git cd survey-sensei/backend conda env create -f environment.yml # Installs all deps (FastAPI, LangChain, etc.) conda activate survey-sensei 2. Configure environment variables: cp .env.local.example .env.local cp .env.local.example .env.local Edit .env.local with your credentials: .env.local OPENAI_API_KEY=sk-proj-... # From platform.openai.com SUPABASE_URL=https://xxxxx.supabase.co # From Supabase dashboard SUPABASE_SERVICE_ROLE_KEY=eyJhbGciOiJIUzI1NiIs... # From Supabase Settings → API RAPID_API_KEY=your_rapidapi_key # Optional: From rapidapi.com OPENAI_API_KEY=sk-proj-... # From platform.openai.com SUPABASE_URL=https://xxxxx.supabase.co # From Supabase dashboard SUPABASE_SERVICE_ROLE_KEY=eyJhbGciOiJIUzI1NiIs... # From Supabase Settings → API RAPID_API_KEY=your_rapidapi_key # Optional: From rapidapi.com 3. Initialize database: python database/init/apply_migrations.py # Applies migrations # Execute SQL code from `backend\database\_combined_migrations.sql` in your supabase project python database/init/apply_migrations.py # Applies migrations # Execute SQL code from `backend\database\_combined_migrations.sql` in your supabase project 4. Start the backend: uvicorn main:app --reload --port 8000 uvicorn main:app --reload --port 8000 Frontend Setup 1. Navigate to frontend: cd survey-sensei/frontend cd survey-sensei/frontend 2. Configure environment variables: cp .env.local.example .env.local cp .env.local.example .env.local Edit .env.local: .env.local NEXT_PUBLIC_SUPABASE_URL=https://xxxxx.supabase.co NEXT_PUBLIC_SUPABASE_ANON_KEY=eyJhbGciOiJIUzI1NiIs... OPENAI_API_KEY=sk-proj-... NEXT_PUBLIC_SUPABASE_URL=https://xxxxx.supabase.co NEXT_PUBLIC_SUPABASE_ANON_KEY=eyJhbGciOiJIUzI1NiIs... OPENAI_API_KEY=sk-proj-... 3. Install dependencies and start dev server: npm install npm run dev # Open http://localhost:3000 npm install npm run dev # Open http://localhost:3000 Testing the Flow Step 1: Submit product and user information Step 1: Submit product and user information Enter an Amazon product URL (must include ASIN) and generate mock data (~3–4 minutes when run locally). ASIN The MockDataOrchestrator builds a realistic e-commerce simulation in your supabase project. For example: Entity Count Composition Purpose Products 11 1 real (RapidAPI) + 6 similar (LLM) + 4 diverse (LLM) Market context for ProductContextAgent Users 13-25 1 main user + 12-24 mock personas (varied ages, locations, purchase patterns) Behavioral diversity for CustomerContextAgent Reviews 30-100+ 10-15 real (RapidAPI) + 20-85 LLM-generated (70% positive, 20% neutral, 10% negative) Signal for ProductContextAgent analysis Transactions 80-170+ Each review → 1 transaction; additional no-review purchases (40% sparsity); 1 "current" delivery (triggers survey) Realistic purchase patterns Embeddings 200-300 All entities → 1536-dim vectors (batch parallel via text-embedding-3-small) Semantic similarity search Entity Count Composition Purpose Products 11 1 real (RapidAPI) + 6 similar (LLM) + 4 diverse (LLM) Market context for ProductContextAgent Users 13-25 1 main user + 12-24 mock personas (varied ages, locations, purchase patterns) Behavioral diversity for CustomerContextAgent Reviews 30-100+ 10-15 real (RapidAPI) + 20-85 LLM-generated (70% positive, 20% neutral, 10% negative) Signal for ProductContextAgent analysis Transactions 80-170+ Each review → 1 transaction; additional no-review purchases (40% sparsity); 1 "current" delivery (triggers survey) Realistic purchase patterns Embeddings 200-300 All entities → 1536-dim vectors (batch parallel via text-embedding-3-small) Semantic similarity search Entity Count Composition Purpose Entity Entity Count Count Composition Composition Purpose Purpose Products 11 1 real (RapidAPI) + 6 similar (LLM) + 4 diverse (LLM) Market context for ProductContextAgent Products Products Products 11 11 1 real (RapidAPI) + 6 similar (LLM) + 4 diverse (LLM) 1 real (RapidAPI) + 6 similar (LLM) + 4 diverse (LLM) Market context for ProductContextAgent Market context for ProductContextAgent Users 13-25 1 main user + 12-24 mock personas (varied ages, locations, purchase patterns) Behavioral diversity for CustomerContextAgent Users Users Users 13-25 13-25 1 main user + 12-24 mock personas (varied ages, locations, purchase patterns) 1 main user + 12-24 mock personas (varied ages, locations, purchase patterns) Behavioral diversity for CustomerContextAgent Behavioral diversity for CustomerContextAgent Reviews 30-100+ 10-15 real (RapidAPI) + 20-85 LLM-generated (70% positive, 20% neutral, 10% negative) Signal for ProductContextAgent analysis Reviews Reviews Reviews 30-100+ 30-100+ 10-15 real (RapidAPI) + 20-85 LLM-generated (70% positive, 20% neutral, 10% negative) 10-15 real (RapidAPI) + 20-85 LLM-generated (70% positive, 20% neutral, 10% negative) Signal for ProductContextAgent analysis Signal for ProductContextAgent analysis Transactions 80-170+ Each review → 1 transaction; additional no-review purchases (40% sparsity); 1 "current" delivery (triggers survey) Realistic purchase patterns Transactions Transactions Transactions 80-170+ 80-170+ Each review → 1 transaction; additional no-review purchases (40% sparsity); 1 "current" delivery (triggers survey) Each review → 1 transaction; additional no-review purchases (40% sparsity); 1 "current" delivery (triggers survey) Realistic purchase patterns Realistic purchase patterns Embeddings 200-300 All entities → 1536-dim vectors (batch parallel via text-embedding-3-small) Semantic similarity search Embeddings Embeddings Embeddings 200-300 200-300 All entities → 1536-dim vectors (batch parallel via text-embedding-3-small) All entities → 1536-dim vectors (batch parallel via text-embedding-3-small) Semantic similarity search Semantic similarity search Step 2: Launch the survey Step 2: Launch the survey Click "Start Survey" and wait 3-5 seconds. What's happening behind the scenes: What's happening behind the scenes: ProductContextAgent → Analyzes reviews with weighted ranking (e.g., recency 50%, quality 40%, diversity 10%) CustomerContextAgent → Profiles purchase behavior and review patterns (e.g., repeat buyer, high review rate, critical tone) SurveyAgent → Combines both contexts to generate initial personalized MCQs (e.g., First set of 3 questions tailored to user+product pair) ProductContextAgent → Analyzes reviews with weighted ranking (e.g., recency 50%, quality 40%, diversity 10%) ProductContextAgent CustomerContextAgent → Profiles purchase behavior and review patterns (e.g., repeat buyer, high review rate, critical tone) CustomerContextAgent SurveyAgent → Combines both contexts to generate initial personalized MCQs (e.g., First set of 3 questions tailored to user+product pair) SurveyAgent Step 3: Answer questions (MCQ-based) Step 3: Answer questions (MCQ-based) Answer 10-12 adaptive MCQ questions. Each response triggers follow-up questions that probe deeper into your concerns (e.g., "battery life" → "how long does it last?" → "does this meet expectations?"). Step 4: Generate and submit review Step 4: Generate and submit review ReviewGenAgent synthesizes your MCQ responses into 2–3 natural language review variations (different star ratings, same sentiment). Pick one, optionally edit, and submit. Real-World Impact Simulations Note: These are projections based on industry benchmarks and reasonable assumptions. Actual results will vary significantly based on implementation, industry vertical, and user behavior. The scenarios below illustrate potential impact, not guaranteed outcomes. Note: These are projections based on industry benchmarks and reasonable assumptions. Actual results will vary significantly based on implementation, industry vertical, and user behavior. The scenarios below illustrate potential impact, not guaranteed outcomes. Scenario 1: Mid-Market E-Commerce Business ($8M revenue) Baseline (Traditional Reviews): Baseline (Traditional Reviews): 50,000 monthly orders → 1,500 reviews → 300 actionable (0.6% of customers) Iteration lag: 4-6 weeks to detect patterns Returns cost: $200k/year 50,000 monthly orders → 1,500 reviews → 300 actionable (0.6% of customers) Iteration lag: 4-6 weeks to detect patterns Returns cost: $200k/year With Survey Sensei (projected): With Survey Sensei (projected): 50,000 orders → 7,500 surveys (15% completion target) → 5,250 actionable (potential 17.5× gain) Iteration speed: Potential to detect issues in Week 1-2 Returns: 2.5% → 1.8% (assumes early issue detection reduces returns) 50,000 orders → 7,500 surveys (15% completion target) → 5,250 actionable (potential 17.5× gain) Iteration speed: Potential to detect issues in Week 1-2 Returns: 2.5% → 1.8% (assumes early issue detection reduces returns) Potential financial impact: Potential financial impact: Returns reduction: $56k/year potential savings Increased repeat purchases: +4% repeat rate → $3.84M/year potential additional revenue Better conversion: +0.5% → $40k/year potential additional revenue Total potential annual benefit: $3.94M/year Returns reduction: $56k/year potential savings Returns reduction: Increased repeat purchases: +4% repeat rate → $3.84M/year potential additional revenue Increased repeat purchases: Better conversion: +0.5% → $40k/year potential additional revenue Better conversion: Total potential annual benefit: $3.94M/year Total potential annual benefit: ROI (assuming full impact realization): ROI (assuming full impact realization): Annual costs: OpenAI API ($180/year) + Infrastructure ($1,440/year) = $1,620/year ROI calculation: ($3.94M - $1.6k) / $1.6k = 243,000% Even at 10% of projected impact, ROI > 24,000% Annual costs: OpenAI API ($180/year) + Infrastructure ($1,440/year) = $1,620/year Annual costs: ROI calculation: ($3.94M - $1.6k) / $1.6k = 243,000% ROI calculation: Even at 10% of projected impact, ROI > 24,000% Scenario 2: Survey-as-a-Service for Small Businesses Potential business model: Potential business model: Target: 250 small e-commerce businesses Pricing: $199/month per business Projected ARR: $597k/year Target: 250 small e-commerce businesses Pricing: $199/month per business Projected ARR: $597k/year Projected unit economics: Projected unit economics: Revenue: $597k/year Estimated costs: $307k/year (API, infrastructure, support, sales) Potential gross profit: $290k/year (48.6% margin) Revenue: $597k/year Estimated costs: $307k/year (API, infrastructure, support, sales) Potential gross profit: $290k/year (48.6% margin) Potential gross profit: $290k/year (48.6% margin) Customer acquisition (estimated): Customer acquisition (estimated): Projected CAC: $500/business Estimated LTV: $199 × 24 months = $4,776 Target LTV:CAC ratio: 9.5:1 (healthy SaaS benchmark: 3:1) Projected CAC: $500/business Estimated LTV: $199 × 24 months = $4,776 Target LTV:CAC ratio: 9.5:1 (healthy SaaS benchmark: 3:1) Target LTV:CAC ratio: 9.5:1 From Ideation to Market Adoption: A Potential 4-Month Journey Month 0: Current MVP State Month 0: Current MVP State Single FastAPI + Supabase + synchronous LLM calls 4-agent system functional with adaptive MCQ generation Cost: $0.10/mock data scenario + $0.002/survey (GPT-4o-mini baseline; batched question generation keeps costs low) Works with mock data; needs real-world validation Single FastAPI + Supabase + synchronous LLM calls 4-agent system functional with adaptive MCQ generation Cost: $0.10/mock data scenario + $0.002/survey (GPT-4o-mini baseline; batched question generation keeps costs low) Works with mock data; needs real-world validation Month 1: Production Hardening + Initial Testing Month 1: Production Hardening + Initial Testing Infrastructure improvements: Infrastructure improvements: Redis session state, background job queues (Celery/BullMQ) Rate limiting, retry logic, circuit breakers Observability: Prometheus + Grafana + Sentry Deploy: Cloud Run + Vercel + Supabase (~$120/month for 10k surveys/day) Redis session state, background job queues (Celery/BullMQ) Rate limiting, retry logic, circuit breakers Observability: Prometheus + Grafana + Sentry Deploy: Cloud Run + Vercel + Supabase (~$120/month for 10k surveys/day) Batch data pipelines: Batch data pipelines: Setup batch context generation: Pre-compute ProductContext and CustomerContext nightly for all active products/users Implementation: Celery/Airflow jobs running nightly, results cached in Redis/PostgreSQL JSONB Benefits: Survey start latency drops from 3-5s to <500ms (contexts already pre-generated) Cost optimization: Generate contexts once daily vs. on-demand per survey (~10× API cost reduction at scale) Setup batch context generation: Pre-compute ProductContext and CustomerContext nightly for all active products/users Setup batch context generation: Implementation: Celery/Airflow jobs running nightly, results cached in Redis/PostgreSQL JSONB Implementation: Benefits: Survey start latency drops from 3-5s to <500ms (contexts already pre-generated) Benefits: Cost optimization: Generate contexts once daily vs. on-demand per survey (~10× API cost reduction at scale) Cost optimization: Early validation: Early validation: Target: 5-10 small Shopify stores (500-2k monthly orders) Deployment: Standalone SaaS with manual CSV upload Goal: 1,000 real surveys → Measure completion rates, agent quality, user satisfaction Key unknowns: Real completion rates vs. mock data, agent consistency across categories Target: 5-10 small Shopify stores (500-2k monthly orders) Deployment: Standalone SaaS with manual CSV upload Goal: 1,000 real surveys → Measure completion rates, agent quality, user satisfaction Key unknowns: Real completion rates vs. mock data, agent consistency across categories Key unknowns: Month 2: Platform Integrations (If Early Metrics Look Promising) Month 2: Platform Integrations (If Early Metrics Look Promising) Service layer architecture: Service layer architecture: # Embedded API integration example @app.route('/webhooks/order_delivered', methods=['POST']) def handle_order_delivered(order_data): response = requests.post('https://api.surveysensei.io/v1/surveys/generate', json={ 'transaction_id': order_data['id'], 'user_id': order_data['customer_id'], 'product_id': order_data['product_id'], 'user_context': {...}, 'product_context': {...} }) survey_url = response.json()['survey_url'] send_email(to=order_data['customer']['email'], body=survey_url) # Embedded API integration example @app.route('/webhooks/order_delivered', methods=['POST']) def handle_order_delivered(order_data): response = requests.post('https://api.surveysensei.io/v1/surveys/generate', json={ 'transaction_id': order_data['id'], 'user_id': order_data['customer_id'], 'product_id': order_data['product_id'], 'user_context': {...}, 'product_context': {...} }) survey_url = response.json()['survey_url'] send_email(to=order_data['customer']['email'], body=survey_url) Initial connectors: Initial connectors: Shopify app: OAuth + webhooks for order.completed events WooCommerce plugin: WordPress plugin triggering post-delivery surveys Zapier connector: No-code integration for quick wins Shopify app: OAuth + webhooks for order.completed events Shopify app: WooCommerce plugin: WordPress plugin triggering post-delivery surveys WooCommerce plugin: Zapier connector: No-code integration for quick wins Zapier connector: Early data patterns (if scale permits): Early data patterns (if scale permits): 5-10k surveys → Begin identifying product category patterns Embeddings infrastructure: Initial indexing of products and users Reality check: Integration complexity often exceeds estimates; OAuth flows require extensive testing 5-10k surveys → Begin identifying product category patterns Embeddings infrastructure: Initial indexing of products and users Reality check: Integration complexity often exceeds estimates; OAuth flows require extensive testing Reality check: Month 3: Analytics Layer + Scale Testing Month 3: Analytics Layer + Scale Testing Basic intelligence features: Basic intelligence features: Admin dashboard: Completion rates, question distribution, sentiment trends Simple alerts: Spike detection for complaint patterns Search interface: Semantic grouping of similar feedback Admin dashboard: Completion rates, question distribution, sentiment trends Admin dashboard: Simple alerts: Spike detection for complaint patterns Simple alerts: Search interface: Semantic grouping of similar feedback Search interface: Scale validation: Scale validation: Test batch processing: Pre-generate contexts nightly Target: 30-50k surveys/day infrastructure capacity Key validation: Can agent quality maintain at scale? Do costs stay predictable? Test batch processing: Pre-generate contexts nightly Target: 30-50k surveys/day infrastructure capacity Key validation: Can agent quality maintain at scale? Do costs stay predictable? Key validation: Market positioning refinement: Market positioning refinement: Focus on validated use cases (likely e-commerce to start) Document what works and what needs improvement Focus on validated use cases (likely e-commerce to start) Document what works and what needs improvement Month 4+: Iterative Improvement Month 4+: Iterative Improvement Production cost economics (estimated for 100k surveys/month): Production cost economics (estimated for 100k surveys/month): Monthly cost: $3,000-5,000 (~$0.03-0.05 per survey) — 92% per-survey APIs, 6% nightly batch contexts, 2% infrastructure Production model considerations: Prototype used GPT-4o-mini ($0.002/survey baseline). Production will likely use latest models (as of writing: GPT-4.5/5, Claude 4 Opus, Gemini 2.0 Ultra at $2.5-6/1M vs. $0.15/1M), raising per-survey costs 15-25× to $0.03-0.05 (batched generation limits cost growth). Monthly cost: $3,000-5,000 (~$0.03-0.05 per survey) — 92% per-survey APIs, 6% nightly batch contexts, 2% infrastructure Monthly cost: Production model considerations: Prototype used GPT-4o-mini ($0.002/survey baseline). Production will likely use latest models (as of writing: GPT-4.5/5, Claude 4 Opus, Gemini 2.0 Ultra at $2.5-6/1M vs. $0.15/1M), raising per-survey costs 15-25× to $0.03-0.05 (batched generation limits cost growth). Production model considerations: Realistic expectations: Realistic expectations: Simulations suggested 10-17× improvement; real-world results will vary Pattern detection speed depends on survey volume and data quality Network effects require significant scale to materialize Simulations suggested 10-17× improvement; real-world results will vary Pattern detection speed depends on survey volume and data quality Network effects require significant scale to materialize Competitive considerations: Competitive considerations: Embeddings tuning: Weeks of optimization work Domain calibration: Requires hundreds of surveys per category Deep integrations: 1-2 months per major platform First-mover advantage exists but isn't insurmountable Embeddings tuning: Weeks of optimization work Domain calibration: Requires hundreds of surveys per category Deep integrations: 1-2 months per major platform First-mover advantage exists but isn't insurmountable What needs ongoing work: What needs ongoing work: Agent prompt tuning based on real user feedback Cost optimization as volume scales Product-category-specific customization Handling edge cases and error modes Begin exploring adjacent verticals based on early feedback Agent prompt tuning based on real user feedback Cost optimization as volume scales Product-category-specific customization Handling edge cases and error modes Begin exploring adjacent verticals based on early feedback Conclusion Survey Sensei demonstrates a practical path toward better customer feedback by combining modern AI capabilities with contextually adaptive data collection: What we built: What we built: 4-agent architecture: ProductContext + CustomerContext + Survey + ReviewGen agents working together Adaptive MCQ generation: Questions that evolve based on answers, not rigid decision trees Vector similarity search: Semantic matching via pgvector for contextual personalization Economic viability: A few cents per survey makes AI-powered feedback broadly accessible across several platforms 4-agent architecture: ProductContext + CustomerContext + Survey + ReviewGen agents working together 4-agent architecture: Adaptive MCQ generation: Questions that evolve based on answers, not rigid decision trees Adaptive MCQ generation: Vector similarity search: Semantic matching via pgvector for contextual personalization Vector similarity search: Economic viability: A few cents per survey makes AI-powered feedback broadly accessible across several platforms Economic viability: Improvements over traditional reviews: Improvements over traditional reviews: Structured feedback through guided MCQ interviews (reduces cognitive load) Real-time pattern detection with agent-based analysis Voice-matched review synthesis (maintains authenticity while saving time) Simulations suggest 10-17× more actionable feedback from the same customer base Structured feedback through guided MCQ interviews (reduces cognitive load) Real-time pattern detection with agent-based analysis Voice-matched review synthesis (maintains authenticity while saving time) Simulations suggest 10-17× more actionable feedback from the same customer base What needs validation: What needs validation: Real-world completion rates with live customers Agent consistency across product categories Long-term ROI at production scale Real-world completion rates with live customers Agent consistency across product categories Long-term ROI at production scale The system works today. Clone the repo, run the setup, and test a survey in 10 minutes. The architecture shows how multi-agent patterns handle complex, context-dependent workflows—not just for surveys, but for any system requiring personalization at scale. If you're building customer feedback systems, recommendation engines, or personalization tools, this architecture offers a concrete reference implementation. The pattern (context gathering → adaptive decision-making → personalized output) generalizes well: The system works today. Customer support triage: Ticket context + User history → Prioritized response Personalized onboarding: Product features + User goals → Custom walkthrough Content recommendations: User preferences + Content embeddings → Ranked suggestions Customer support triage: Ticket context + User history → Prioritized response Customer support triage: Personalized onboarding: Product features + User goals → Custom walkthrough Personalized onboarding: Content recommendations: User preferences + Content embeddings → Ranked suggestions Content recommendations: The shift toward specialized agents collaborating on tasks represents a practical middle ground between monolithic models and over-engineered microservices. It's early, but the economics and technical patterns are sound enough to build on. References and Further Reading Questions? Ideas? Feedback? Questions? Ideas? Feedback? GitHub: github.com/arnavvj/survey-sensei Contributions are welcome—feel free to open an issue or leave a star ⭐ GitHub: github.com/arnavvj/survey-sensei github.com/arnavvj/survey-sensei Contributions are welcome—feel free to open an issue or leave a star ⭐ Academic Papers Academic Papers Automated Survey Collection with LLM-based Conversational Agents (arXiv, 2024) - Framework for phone-based surveys using conversational LLMs with 98% extraction accuracy Embedding in Recommender Systems: A Survey (arXiv, 2023) - Comprehensive survey on embedding techniques for recommender systems at scale Unified embedding: Battle-tested feature representations for Web-scale ML systems (Google Research, NIPS '23) - Real-world deployment of embeddings in production systems Automated Survey Collection with LLM-based Conversational Agents (arXiv, 2024) - Framework for phone-based surveys using conversational LLMs with 98% extraction accuracy Automated Survey Collection with LLM-based Conversational Agents Embedding in Recommender Systems: A Survey (arXiv, 2023) - Comprehensive survey on embedding techniques for recommender systems at scale Embedding in Recommender Systems: A Survey Unified embedding: Battle-tested feature representations for Web-scale ML systems (Google Research, NIPS '23) - Real-world deployment of embeddings in production systems Unified embedding: Battle-tested feature representations for Web-scale ML systems Industry Reports Industry Reports Global Study: How Consumers Share Feedback, 2025 (Qualtrics XM Institute) - Analysis of 23,000+ consumers showing direct feedback declining while indirect feedback increased 60% The State of Customer Experience Management, 2025 (Qualtrics XM Institute) - Annual CX management trends and priorities Power forward: Five make-or-break truths about next-gen e-commerce (McKinsey, 2024) - E-commerce trends with focus on AI and generative AI adoption The agentic commerce opportunity (McKinsey, 2024) - How AI agents are transforming consumer shopping experiences Global Study: How Consumers Share Feedback, 2025 (Qualtrics XM Institute) - Analysis of 23,000+ consumers showing direct feedback declining while indirect feedback increased 60% Global Study: How Consumers Share Feedback, 2025 The State of Customer Experience Management, 2025 (Qualtrics XM Institute) - Annual CX management trends and priorities The State of Customer Experience Management, 2025 Power forward: Five make-or-break truths about next-gen e-commerce (McKinsey, 2024) - E-commerce trends with focus on AI and generative AI adoption Power forward: Five make-or-break truths about next-gen e-commerce The agentic commerce opportunity (McKinsey, 2024) - How AI agents are transforming consumer shopping experiences The agentic commerce opportunity Multi-Agent Systems Resources Multi-Agent Systems Resources LangGraph: Multi-Agent Workflows (LangChain Blog, 2024) - Building multi-agent systems with LangGraph Multi-agent network tutorial (LangChain Documentation) - Step-by-step guide to multi-agent collaboration patterns LangGraph: Multi-Agent Workflows (LangChain Blog, 2024) - Building multi-agent systems with LangGraph LangGraph: Multi-Agent Workflows Multi-agent network tutorial (LangChain Documentation) - Step-by-step guide to multi-agent collaboration patterns Multi-agent network tutorial Technical Documentation Technical Documentation LangChain: python.langchain.com LangGraph: langchain-ai.github.io/langgraph pgvector: github.com/pgvector/pgvector OpenAI Embeddings: platform.openai.com/docs/guides/embeddings LangChain: python.langchain.com python.langchain.com LangGraph: langchain-ai.github.io/langgraph langchain-ai.github.io/langgraph pgvector: github.com/pgvector/pgvector github.com/pgvector/pgvector OpenAI Embeddings: platform.openai.com/docs/guides/embeddings platform.openai.com/docs/guides/embeddings All code examples and architecture diagrams are from the Survey Sensei codebase. All code examples and architecture diagrams are from the Survey Sensei codebase. All code examples and architecture diagrams are from the Survey Sensei codebase.