InsureBot

Inspiration

Life is very uncertain and you may find yourself in tricky situations wondering if you are covered through your insurance. Policy coverage documents are often complicated and difficult to decipher. Would it not be handy if there is someone who can answer any and all of your queries 24*7? That's why we cam e up with the idea of InsureBot.

What it does

Our InsureBot provides users with a simple interface to unlock the potential of exploring their policies and coverage. Whether you are a novice or an insurance expert, our bot got all the answers for you. It answers the users queries about all that is covered by their insurance based on the chosen policy

How we built it

InsureBot is a retrieval QA chain which is built on langchain and llama2 integrated with a telegram bot. The full policy documents are embedded into a vector database, based on the policy chosen by the user, relevant chunks of information is retrieved from the DB and passed onto the llama2 chat model to have human like conversations regarding their insurance coverage.

Challenges we ran into

1. AWS Sagemaker did not work due to incompatibility and location mismatch issues.
2. Compute resources shortage.
3. Difficulty in creation of a vector database using chromadb due to several incompatible libraries and packages.

Accomplishments that we're proud o

1. Efficient utilization of LLM chat capabilities by providing the relevant context rather than fine tuning the entire model to fit the use case.
2. Leveraging the power of vector databases to store the policy information given the ever growing data to form a easy retrieval of relevant information.

What we learned

1. Lots of patience, team work and perseverance.
2. Deeper understanding of LLMs and langchain framework.

What's next for InsureBot

Based on the database, it is capable of being a knowledge bot in various contexts to serve customer needs.

#Zurich Insurance

Built With

• chromadb
• gradio
• huggingface
• langchain
• llama2
• python