Skylark Labs Secures $21M Deal to Deploy Self-Aware AI Across 6,000 Police Systems

Written by jonstojanjournalist | Published 2025/09/17
Tech Story Tags: skylark-labs-ai | self-aware-ai-systems | traffic-enforcement-ai | ai-accuracy-decay | dr.-amarjot-singh | adaptive-ai-technology | kepler-ai-platform | good-company

TLDRTraditional AI silently degrades over time, missing violations and risking safety. Skylark Labs, founded by Dr. Amarjot Singh, has landed a $21M deal to deploy its Kepler™ self-aware AI across 6,000 police systems. Unlike static AI, Kepler monitors its own performance, adapts locally, and retrains on the fly—cutting costs, boosting accuracy, and setting a new standard for resilient AI.via the TL;DR App

The startup’s deployment across 6,000 law enforcement assets aims to address the invisible decay in AI accuracy without needing costly retraining or downtime.


Across thousands of cities, artificial intelligence (AI) systems are quietly failing, and no one knows it. Over time, traffic AI misses more violations, loses accuracy, and stops working as intended. The reason? Changes in the environment or user behavior that the system was never trained to handle.

Skylark Labs, a startup founded by Dr. Amarjot Singh in New York, has secured a $21 million, three-year contract to deploy its Kepler™ platform across 6,000 police systems in a leading Asian country. What sets Kepler™ apart is its self-awareness. While most AI silently degrades, Kepler is designed to continuously check its performance and adapt on the fly. It doesn’t just work today; it can get better over time.

“The dirty secret of AI is that most systems start failing the day they go live,” Dr. Singh says. “They can’t handle new traffic patterns, evolving vehicle designs, or changes in driver behavior. Worst of all, no one notices until it’s too late. We built Kepler™ to fix that.”

The Silent Degradation Crisis

Most traditional AI systems are trained once on past data and then deployed as if the world will never change. But in traffic enforcement, that assumption fails fast.

Roads evolve, new car models hit the streets, driver behavior shifts, and weather and lighting change. But the AI stays stuck, still trying to make decisions based on outdated data.

“According to industry estimates, local governments spend millions annually on AI maintenance without clear visibility into performance degradation—creating hidden operational risks and unstable revenue projections,” Dr. Singh adds. “Traditional AI doesn’t even realize it’s failing, so it never raises a flag. It just keeps getting worse in the background.”

Many AI systems follow an expensive and frustrating pattern: they work well at first, then slowly get worse. To fix them, cities have to pay significant capital for retraining or replacing the system entirely, again and again.

Dr. Singh emphasizes, “The impact of this lapse is concerning. They can lose revenue from missed violations and, more importantly, put public safety at risk because their AI can’t keep up with change.”

AI That Knows When it’s Falling Behind

Skylark Labs’ innovation didn’t happen overnight. It was born from years of research under the DARPA Lifelong Learning Machines program, where founder Dr. Amarjot Singh worked on AI that doesn’t just process data; it watches itself, knows when it’s confused, and adapts in real time to stay accurate.

“Our biggest innovation is self-awareness,” Dr. Singh explains. “The AI doesn’t just detect traffic violations; it detects when it might be missing them.”

Dr. Singh explains how it works: Every patrol car using Skylark Labs’ Kepler™ platform runs multiple AI models at once. The main model flags violations like speeding or wrong turns. But behind the scenes, a second layer of models constantly monitors performance, watching for signs the system might be missing something new, like a vehicle type it’s never seen before or an unusual traffic pattern.

When it spots a potential gap, Kepler is designed to automatically adjust its internal settings and retrain itself locally, so no internet or human engineers are needed. With this self-aware and memory-efficient setup, Skylark Labs’ AI keeps learning in the field, avoiding expensive updates, and getting smarter with every deployment.

Beyond Traffic: Universal Applications

The self-aware AI principle extends beyond traffic enforcement to any application where conditions change over time.

For military applications, Skylark Labs’ Tracer AI software can scan aircraft carrier decks for foreign object debris. As new aircraft types and deck configurations are introduced, the system can adapt automatically.

Skylark Labs’ Scout towers were deployed at Indiana's Department of Transportation, which creates a network that adapts to changing traffic flows, seasonal patterns, and infrastructure modifications.

The same learning-adaptive principle could be applied to rapidly changing domains like healthcare, border security, disaster response, and financial fraud detection—where conditions evolve faster than centralized updates can keep pace.

The Competitive Void

While some firms have built impressive AI capabilities, they often share the same critical limitation: their systems operate blindly. They can detect, classify, and respond, but they lack the capacity to assess their performance. When conditions shift, they fail quietly. And worse, they don’t know it.

“The industry has obsessed over making AI faster, sharper, and more efficient,” argues Dr. Singh. “But no one has addressed the foundational flaw: AI that doesn’t know when it’s wrong. That’s not just a technical gap; it’s a trust crisis.”

Where traditional AI platforms rely on human audits, reactive updates, and expensive retraining to maintain performance, Skylark Labs’ self-aware systems operate in a different paradigm. Without external intervention, they can continuously monitor their confidence, detect anomalies in real time, and adapt on the edge. It represents a significant advancement in how AI is built, deployed, and trusted at scale.

A Smarter AI That Pays for Itself

Skylark Labs’ self-aware AI offers cost savings and potential long-term value. By focusing on sustained performance, it can contribute to consistent revenue, potentially lower update costs, and enhance public safety through better violation detection. Cities are less likely to encounter sudden performance drops or unexpected maintenance charges.

“It’s like hiring a traffic officer who gets better with experience, not worse,” Dr. Singh says. “That’s a completely different kind of value.”

As more AI systems are deployed worldwide, the problem of silent failure will only grow. Most systems trained on today’s data may fall behind as conditions change, but Skylark Labs’ approach flips the script.

“The future belongs to AI that can learn and evolve,” Dr. Singh adds. “Static AI is dead AI; it just doesn’t know it yet.”

With its $21M deployment in Asia, Skylark Labs is proving that self-aware AI can handle real-world scale. And if it succeeds, it won’t just set a new standard for traffic enforcement. It will reshape expectations for AI in every sector where the world keeps changing.

“We’re not just building better AI,” Dr. Singh concludes. “We’re building AI that knows how to stay better.”

As governments and enterprises increasingly demand resilient AI solutions with minimal overhead, Skylark’s approach could signal a shift in how foundational AI infrastructure is procured, deployed, and measured. Who knows, the silent AI failure crisis may finally be coming to an end.


VentureBeat newsroom and editorial staff were not involved in the creation of this content.


Written by jonstojanjournalist | Jon Stojan is a professional writer based in Wisconsin committed to delivering diverse and exceptional content..
Published by HackerNoon on 2025/09/17