BeAwareOfBanks
AI that reads a loan agreement and tells the borrower what they just signed.
The problem
A loan in India comes with 40 to 80 pages of legal English. Almost nobody reads all of it, and the clauses that hurt two years later are the ones buried deepest: rates that move without notice, aggressive recovery rights, cross-default triggers that link one loan to another.
What we built
The product takes an uploaded agreement, asks the borrower a few questions, and returns a risk report in about three minutes. The wider Evodart team shaped and delivered it. My work sat underneath, on the AI layer: strengthening the LLM and RAG integrations, tuning the pipelines that pull the right clauses into context, supporting model-security practices, and fixing product issues as they surfaced.
- Document deep-scan that flags dangerous clauses and explains each one in ordinary language
- Guardian Advisor chatbot, which answers questions about the user's own uploaded agreement
- RAG pipelines tuned for retrieval accuracy across long legal documents
- Model-security work and bug fixes alongside the Evodart product team
- Live in beta at beawareofbanks.com
My contribution
What I owned on the team
AI Engineer Intern, 11 May to 30 June 2026. I contributed to the LLM and RAG integrations, refined the AI pipelines, worked on model-security practices, debugged the product, and coordinated delivery with the team through ClickUp.
How it turned out
The platform is live and taking real documents. What I contributed is the retrieval and generation layer that turns a dense contract into something a borrower can act on before signing.