FinSight AI

ChatGPT for Finance, but with real documents + structured insights.

Overview

FinOps AI is an end-to-end system that analyzes financial reports, earnings PDFs, and news articles, and produces structured insights using a Retrieval-Augmented Generation (RAG) pipeline.

It extracts text → embeds it → stores vectors in FAISS → retrieves relevant chunks → feeds them into an LLM → outputs risk summaries, sentiment, key metrics, and actionable insights.

This project demonstrates:

1. Python LLM engineering
2. RAG architecture
3. Embeddings + FAISS
4. Structured LLM output
5. End-to-end product thinking

Features (MVP)

1. Upload PDF, URL, or plain text
2. Automatic text preprocessing + chunking
3. Embedding generation (OpenAI / Jina / Cohere)
4. Vector indexing using FAISS
5. RAG query pipeline
6. LLM generates:
   • risk factors
   • sentiment analysis
   • opportunities & threats
   • extracted metrics
   • structured JSON summary
7. REST API (FastAPI)
8. Ready for React UI integration

Architecture Diagram

High-Level Architecture:

API Endpoints

POST /analyze
Analyzes a financial document or URL.

Request:

{
  "type": "pdf",
  "input": "http://newslink.com/article123",
  "analysis-type": "comprehensive-review",
  "focus-area": "risk-&-revenue",
  "query": "summarize risks and metrics"
}

Response (simplified):

{
  "summary": [
    "Bullet point 1",
    "Bullet point 2",
    "Bullet point 3"
  ],
  "sentiment": "neutral",
  "risk_factors": ["..."],
  "opportunities": ["..."],
  "key_metrics": {
    "revenue": { "value": "$4.2B", "change": "15%", "direction": "up" },
    "eps": { "value": "$1.24", "change": "8%", "direction": "up" },
    "op_margin": { "value": "22.5%", "change": "1.2%", "direction": "down" },
    "free_cash_flow": { "value": "$850M", "change": null, "direction": "unknown" }
  },
  "confidence_score": 87.5,
  "sources_used": 5,
  "citations_used": 3
}