Hackasaurus Rex

Inspiration

Inspirartion for this idea came from a recent personal experience. A family member was recently diagnosed with a medical condition that requires surgery. Insurance company automated chatbots were not very helpful and calling the company everyday became very time-consuming because we had to give context to the agent everytime we called. And we had new questions everyday. So this is our attempt to simplify the process.

What it does

This allows users to interact with their insurance policy in a new way. We built a chatbot that can handle complicated questions from users. It allows users to ask for information not even explicitly mentioned in their insurance policy. For example, a user can ask "Is ASD procedure covered in my insurance? If so, what is my total out-of-pocket costs?" The tool will first figure out that ASD is a short-form for Atrial Septal Defect (a heart condition). Then it will understand that is a required procedure and hence covered under xyz conditions mentioned in the policy. Then it should calculate average cost for different types of procedures options available and out-of-pocket for the user.

How we built it

The chatbot's prompt chain follows a specific structure. User input is processed to identify the use cases to address. Series of prompts were designed to extract relevant information from the input, utilize search tools for additional context on medical procedures, classify the input into fixed categories, retrieve relevant document embeddings, and generate an answer using an LLM model based on the obtained context. Currently, the chatbot is restricted to addressing a specific use case. However, with the assistance of domain experts, it can be expanded to cover a broader range of use cases. This would reduce customer service time and effort while providing users with a powerful tool to interact with complex insurance PDFs instead of reading them.

Challenges we ran into

One major challenge faced was integrating custom chains within the conversation chain. This issue remained unresolved, indicating difficulties in seamlessly incorporating custom functions into the chatbot. Limitations: While langchain offers various use cases and customization options, there were limitations in having absolute control over integration. Combining different elements often led to failures, indicating the challenge of customizing various components of langchain.

Accomplishments that we're proud of

We're proud of learning all different capabilities of vector databases and langchain. Making a poc like this within 1 week was a daunting task, however, we're proud to have a functional poc. We learned that for accurate conversational chain, we had modify our approach several times and that just fine as long as we're failing forward.

What we learned

The vastness and complexity of langchain functions and their integration were overwhelming but intriguing. Understanding how each tool/function works and integrates with others was a significant learning experience.

What's next for Hackasaurus Rex

Continue developing the product and include several other use-cases. As mentioned above, the chatbot is restricted to addressing a specific use case. However, with the assistance of domain experts, it can be expanded to cover a broader range of use cases. This would reduce customer service time and effort while providing users with a powerful tool to interact with complex insurance PDFs instead of reading them.