Inspiration
Financial due diligence is still a slow and manual process. Investors, analysts, and finance teams spend hours reading annual reports, earnings transcripts, and financial statements to understand a company's performance and identify potential risks. As someone interested in applying AI to financial intelligence, I wanted to explore whether multiple AI agents could collaborate to automate this workflow while keeping the reasoning transparent and explainable. This led to FinSight AI, a multi-agent financial intelligence platform that helps users analyze financial documents, identify risks, and generate structured investment insights.
What it does
FinSight AI uses specialized AI agents that work together to perform different stages of financial analysis. The platform can: Analyze financial reports and earnings documents Retrieve relevant information using RAG Extract key financial metrics Identify business and financial risks Generate executive summaries Produce evidence-backed investment insights Instead of relying on a single LLM prompt, FinSight AI breaks the workflow into multiple specialized agents that collaborate to complete complex financial analysis.
How we built it
The application consists of: Frontend: Streamlit + Python Backend: FastAPI (Python) LLMs: Gemini/OpenAI, Qwen LLM model Agent Orchestration: LangGraph Retrieval: RAG with vector embeddings Vector Database: ChromaDB Embeddings: Sentence Transformers Deployment: Still in localhost The workflow combines document retrieval, semantic search, AI reasoning, and multi-agent collaboration to generate explainable financial insights.
Challenges we ran into
Some of the biggest challenges included: Designing a reliable multi-agent workflow instead of a single-agent pipeline. Improving retrieval quality for financial documents. Reducing hallucinations through RAG and structured prompts. Coordinating multiple AI agents while maintaining context. Building a workflow that remains explainable instead of acting like a black box.
Accomplishments that we're proud of
Built a working multi-agent financial analysis platform. Integrated AI agents with retrieval pipelines. Created an end-to-end FastAPI backend and React frontend. Designed a modular architecture that can easily support additional financial workflows.
What we learned
This project reinforced that production AI systems require much more than calling an LLM API. Reliable AI applications need: High-quality retrieval Structured workflows Agent orchestration Evaluation Explainability Robust backend engineering Building FinSight AI gave us practical experience designing AI systems that move beyond simple chat interfaces toward real-world financial workflows.
What's next for FinSight AI
Future improvements include:
Support for SEC filings and earnings-call transcripts Real-time financial market data Portfolio risk analysis Financial anomaly detection AI-powered due diligence reports Evaluation framework for agent quality Multi-user collaboration and enterprise deployment Bring it in production-live
Log in or sign up for Devpost to join the conversation.