Building Automation That Works: Lessons from MVPs with LLMs by Aparimeya Taneja

Emerging technologies such as Artificial Intelligence (AI) are set to only rise in prominence as the decade continues. As a result, businesses are moving to integrated systems that reduce the menial tasks people face in the workplace. Aparimeya Taneja, a developer and engineering expert, has spent his career building solutions both at scale and at startups. Drawing on his experience at DoorDash and Fulcrum Tech, he highlights the practical challenges and opportunities of scaling automation from MVPs to fully deployed agentic solutions.

Introducing Tech Expert Aparimeya Taneja

In high school, Aparimeya showed promise in STEM, leading him to pursue a career in engineering. Aparimeya would then officially begin his engineering career after graduating early from Duke University. Shortly after completing his studies, he would join DoorDash’s new verticals organization as a part of the search experience team.

Aparimeya would spend a year in this team before transitioning to the new verticals Machine Learning (ML) team, where he honed his technical skills and contributed to an automation project for over six months before moving to the new verticals catalog team for another six months.

After gaining experience from working at such a large organization, he joined Paraform, a startup focused on recruitment solutions for industries like software development and financial services.

He then joined Fulcrum as a Founding Engineer to apply his automation expertise to the insurance industry. Fulcrum offers secure AI products that allow brokerages to scale their accounts without headcount. Aparimeya’s prioritization of automation solutions has supported the tech company's continued growth.

Building MVPs for Automation with LLMs

During his time at DoorDash, Aparimeya led efforts to develop MVPs for automation of tasks that were outsourced to BPOs. His contributions led to an opportunity to collaborate with the company’s ML and Automation team. He saw the potential of LLMs to automate manual outsourced work that many organizations typically spend millions of dollars on. DoorDash enabled him to work on LLM-powered projects to streamline various business processes.

Aparimeya was the first developer at DoorDash to propose using automation, earning noteworthy mention for his efforts. Using LLMs, he worked with a team of like-minded experts to build product knowledge for the company.

His work included building product knowledge systems that improved personalization, selection, and fulfillment. By addressing challenges like incomplete data and inferential reasoning, Aparimeya helped optimize DoorDash’s operations.

While he was instrumental in many of these projects, Aparimeya experienced many challenges moving MVPs forward in a large organization like DoorDash. After pushing to build automation solutions for the company and working on a few MVPs, he was required to return to regular projects.

The constraints of a large organization limited his ability to see these projects through to completion. This helped Aparimeya realize he wanted to work in an early-stage startup, which has seen him pivot to Fulcrum Tech, where he spearheads projects with the company’s technical staff.

At Fulcrum, issues such as data handling, model training, and scalability began to arise. These problems were often intricate and technically challenging, even for an expert like Aparimeya. Despite these hurdles, he worked tirelessly on automating the manual processes that plague brokers, making them more adaptable while cutting costs and reducing time taken for tasks from hours to minutes.

Best Practices for Building Scalable Automation Solutions

Aparimeya’s work at DoorDash and Fulcrum has empowered business processes through automation solutions. He notes that scaling from MVPs to real-world applications allows brokerages to balance costs, increase efficiency, and enhance user adoption, ultimately streamlining their workflows.

Drawing from his experience, Aparimeya advocates for a strategic approach to building scalable automation solutions. Key best practices for him include:

Balancing Cost vs Value Provided

In Aparimeya’s experience, automation solutions are usually priced per outcome / output to end customers, thus it is important to make sure your cost of operation for running automations does not have negative per unit economics.

The value generated for your customer is the difference between what it would have cost them in time, quality and expenses to reach the same output without your automation vs with your solution. Thus, you should always have a clear idea of how you are creating value for yourself and your customers when building AI products.

Storing Unstructured Data at Scale

Aparimeya emphasizes that while thinking about data storage, the key objects to solve for are:

1. Robust data storage systems with redundancy and infrastructure to support scalability. This ensures businesses can securely access, protect, and recover information. Enterprise grade SQL services are a great solution to get started.

   
   

2. Enriched knowledge graph to build features and workflows on top off. This will depend on the domain you are building solutions for and requires an in-depth understanding of how your users do their workflows manually today. Generally, it is always more useful to extract key datapoints and store them in a way that they are easy to index instead of dumping all data as storage objects or JSON blobs.

Evaluating User Adoption

Throughout his work, Aparimeya recognized that user adoption was another critical component for implementing any successful automated processes. Initiatives such as training programs are essential in equipping users with the ability to operate upgraded processes confidently and could further promote automation solutions' functionality. Businesses can balance user adoption by gathering feedback and customizing user interfaces accordingly while potentially integrating third-party applications. In the case of Fulcrum, Aparimeya and his team integrate directly with their customers’ AMS ( Agency Management System).

Another effective way to quantitatively access user engagement is through telemetry through your application that records user sessions, as well as to generate usage statistics to observe trends in usage over time as you ship more features.

Maintaining Speed

While automation systems can streamline work processes, they are often prone to overloading, and due to a lack of visibility can appear slow to end users. Thus, it is important to try to parallelize as many tasks as possible in your end to end automations and use robust queues for handling long running jobs. You can also boost speed by reducing duplicate work and provisioning more resources.

Aparimeya believes in designing an application architecture such that the bottleneck of the system is not at any one critical point and that resources are neither idling  nor always overloaded.

Observing and Optimizing

Automation systems require constant monitoring to make sure all processes operate as expected. Supervising automated workflows can allow businesses to easily identify components that are a bottleneck to the whole process, as well as know the health of all the various microservices at any given moment. When a company monitors its system’s performance regularly, it can develop ways to optimize long-term automation processes.

Aparimeya advocates that this kind of monitoring should also extend to the final output your users see so that you are always iterating on the user experience.

Product Feedback Loop

Shipping features and automations is only half of what a good solution platform should do. Every time a new feature or workflow is released, apart from internal testing it should be rolled out to select users at first. Then, detailed feedback should be collected from this pilot group on the basis of which future iterations should be improved. It is important to maintain your product roadmap however user feedback should always be considered when deciding what is the most important thing to work on.

Aparimeya Taneja’s Advice for Engineers Entering the Automation Space

Aparimeya Taneja encourages aspiring engineers to develop a solid foundation in computing or engineering as they step into the world of automation technology. He emphasizes that strong programming and software development skills can greatly enhance success. Familiarity with the domain of what you will be automating is also essential, as that knowledge will help you deliver deeply technical solutions that provide a lot more value than just a chat bot.

In addition to technical skills, engineers should prioritize User Interface (UI) and User Experience (UX) design when creating automation solutions. Ensuring accessibility and ease of use can improve the effectiveness and adoption of these tools. Aparimeya also highlights the importance of creative thinking in implementing automation systems to maximize their potential.

Looking ahead, Aparimeya Taneja envisions a future where automation is seamlessly integrated into business processes, reducing repetitive tasks for human workers. By sharing his expertise, he hopes to make automation a standard practice across industries in the future.