The problem Talk The Block solves

Background

Blockchain data is heavily structured and optimized for machine readability. It's impressingly difficult to read these millions of data points and make any sense. It's time to take help from Conversational AI technologies such as LLM (Large Language Model) and make the best out of the data available for answering some questions users may have.

Solution

The solution is to buid an LLM app that can answer questions posed by users in some detail.

Challenges I ran into

While building an LLM app to answer blockchain transactional data questions naturally, it is super tough to ensure that the data is pretty accurate and on spot. So, this is a major challenge in general and termed "Hallucination" in the AI space.

This issue has more direct impact in the Finance Space as we don't want an app to "hallucinate" on our wallet balance, or on some other key insights needed by the user.

The approach to closely track the validity of the app is to use LangSmith.