Inspiration

As someone who invests heavily in stocks and tracks market trends, I wanted to explore whether there are real ways to get ahead of the market. Most market-moving information starts in SEC filings, but by the time it reaches the news, the opportunity is often gone.

What it does

FilingPulse watches SEC EDGAR in real time and flags market-moving filings the moment they post. It focuses on 8-Ks, insider purchases (Form 4s), and activist stakes (13Ds). A local AI model reads each filing, decides whether it leans bullish or bearish, and pushes an alert to a Robinhood-style dashboard. Each alert shows a plain-English summary, the key quote from the filing, a confidence score, how much the stock has already moved, and whether the company issued a press release, a strong hint that the news hasn't caught up yet.

How we built it Backend: Python and FastAPI poll EDGAR's live feed every 30 seconds, respecting the SEC's rate limits, and map each filing to its stock ticker. Parsing: 8-Ks are split by item number, and Form 4 insider trades are parsed directly from their XML, no AI needed. AI scoring: a small open-source Qwen model runs locally through Ollama, returning structured JSON with direction, confidence, and evidence. Running locally keeps it free and private. Prices: yfinance adds live prices, market cap, and the % move since each filing. Frontend: React and Tailwind, with live alerts streamed through Server-Sent Events. Challenges we ran into

We quickly realized that competing with hedge funds and institutional traders on speed is impossible: their algorithms read filings in milliseconds. So we changed the goal. Instead of beating Wall Street, FilingPulse aims to beat the news cycle and get information to everyday investors before it shows up in headlines. That meant learning which filings actually go uncovered, like small-cap 8-Ks with no press release and quiet insider buying. Getting a small local model to give consistent, structured answers on dense legal text was another challenge.

Accomplishments that we're proud of An end-to-end pipeline that goes from a raw SEC filing to a scored, readable alert in near real time. Running the AI entirely locally, with no API costs. Features built around our "beat the news" idea: filing-to-alert latency, "no press release" flags, and a check on whether the price has already moved. A clean, intuitive dashboard that makes dense filings readable at a glance.

Built With

  • apscheduler-data:-sec-edgar-atom-feed
  • backend:-python
  • fastapi
  • lightweight-charts
  • makefile
  • ollama-(local-llm)
  • sqlmodel/sqlite
  • sse-ops:-.env-config
  • tailwind
  • vite
  • yfinance-frontend:-react
Share this project:

Updates

Submission history