FinSight

Inspiration Investors have access to more financial information than ever, but understanding why a stock moved still requires searching through charts, news articles, company filings, and market data. We wanted to bring those pieces together in one place. That idea became FinSight, a platform that helps everyday investors understand what happened to a stock, why it likely happened, and what followed similar movements in the past. What it does Users can search for any publicly traded stock and investigate either its latest trading session or its most unusual recent movement. FinSight then walks the user through a clear investigation:

• How unusual was the movement?
• Was it caused by the company, its industry, or the broader market?
• What news and SEC filings may explain it?
• Has a similar movement happened before?
• What happened to the stock afterward? The evidence timeline brings together reporting from The New York Times and Yahoo Finance alongside SEC filings. Gemini reviews these sources and identifies the evidence most relevant to the movement. FinSight also includes an ElevenLabs audio briefing, allowing users to listen to the investigation instead of reading the entire report. How we built it We built the frontend with React and Vite and created the backend with Python and Flask. The application is hosted on DigitalOcean. FinSight uses Yahoo Finance for market data, The New York Times and Yahoo Finance for news, and SEC EDGAR for company filings. We use Google Gemini to review and organize the collected evidence, Tiger Data to support financial and time-series data, and ElevenLabs to generate the audio briefing. Our own backend performs the core market analysis. Gemini helps interpret supporting evidence, but it does not decide the numerical results. FinSight also checks when each source was published so that information released after the stock moved is not presented as its cause. Challenges we faced Our biggest challenge was creating explanations that were useful without presenting guesses as facts. A news article published near a stock movement may be related, but timing alone does not prove that it caused the movement. To address this, we combined market calculations, publication times, AI-assisted evidence review, and direct links to the original sources. This allows users to understand where each conclusion came from. We also faced API limits and inconsistent results across large date ranges. We added caching, divided longer searches into smaller time periods, and loaded the timeline progressively. This reduced unnecessary requests and helped us provide broader evidence coverage. Another challenge was making the application reliable when an outside service became unavailable. FinSight can still display its market analysis if an AI, news, or audio service fails. Any evidence that has not been reviewed is clearly labeled. What we learned We learned that responsible financial AI needs transparency. An answer is much more useful when users can see the calculations, timing, and sources behind it. We also learned how to combine several APIs into one reliable experience while managing request limits, incomplete data, and service failures. Most importantly, we learned that AI works best here as an evidence-reviewing tool supported by real market analysis. What’s next We would like to add portfolio-level investigations, personalized watchlists, risk alerts, and automatic notifications when unusual market activity occurs. Our goal is to make financial research easier to understand without removing the evidence behind it. FinSight helps investors move from “What happened?” to “Now I understand why.”

Built With

• ai/ml
• css3
• digitalocean
• docker
• elevenlabs
• flask
• gemini
• github
• gunicorn
• html5
• javascript
• natural-language-processing
• newyorktimesapi
• postgresql
• python
• react
• restapi
• secedgar
• tigerdata
• timescaledb
• vite
• yahoofinance