2025Automation

Financial RAG chatbot

Query financial reports in plain language, with answers you can verify.

Context: A personal project born from a recurring need: pulling one precise figure out of a company annual report.
Problem: A financial report runs to hundreds of pages. A language model asked about it directly invents figures with complete confidence, which is worse than no answer at all: the error goes unnoticed.
Approach: A RAG pipeline orchestrated in n8n: document chunking, embedding generation, vector storage, then retrieval of the relevant passages injected as context before the model call. DeepSeek for generation, LangChain for the retrieval chain, a webhook as the entry point.
Outcome: Every answer ships with the source passage it came from. Checking one takes seconds, which beats taking it on trust.