FinSight AI

FinSight AI is an enterprise-grade, multi-tenant financial intelligence platform built for chartered accountants, non-technical finance stakeholders and business users to autonomously extract, analyze, and generate actionable insights from bank statements through conversational AI and visual analytics dashboards—turning static PDFs into structured data, KPI intelligence, anomaly signals, and citation-backed answers without requiring spreadsheets, SQL, or analyst support.

What we built / implemented

• Built an automated PDF → structured data pipeline using advanced table detection and regex-based parsing to extract account profiles, transaction metadata, and statement periods with robust multi-format date handling.
• Developed an intelligent transaction type inference engine using narration keyword analysis, amount sign detection, and channel heuristics to accurately classify debit/credit flows across diverse banks and statement layouts.

• Implemented a vector search architecture combining Azure AI Services with MongoDB Atlas Vector Search for semantic retrieval over parsed transaction documents.
• Built a LangChain-orchestrated RAG pipeline enabling context-aware, citation-backed Q&A over user documents with temperature-controlled LLM responses (GPT-4).
• Added recursive character-level chunking with metadata enrichment to improve retrieval precision and maintain source traceability.

• Implemented metadata persistence for audit trails and document ownership verification.

• Built real-time financial visualizations using Recharts (bar/line/pie dashboards for transaction analysis, cash flow, channel distribution).