Hackathon theme: Creative manipulation of data

Yield takes raw financial news and transforms it through NLP and quantitative finance into personalised, actionable portfolio signals — turning unstructured text into position-sizing guidance backed by the Kelly Criterion.

1. Browser extension — Browse to any financial news article. The extension scrapes the title and body, sends it to the local engine, and the dashboard updates in real-time.

2. NLP engine (FinBERT + sentence embeddings) — The article text is analysed for sentiment (positive / negative / neutral) and matched against a semantic theme bank and company bank to detect which of your holdings are materially affected.

3. Kelly Criterion position sizing — For each impacted holding the engine computes a Half-Kelly fraction based on the detected sentiment and confidence score, then translates that fraction into a concrete buy / sell / hold recommendation expressed as an allocation percentage and an implied dollar trade size.

4. React dashboard — A SPA served by the FastAPI backend with three views:

   • Analysis — per-holding signal cards with justification, confidence, Kelly guidance, implied trade size, and evidence quotes. Stores and displays the last 3 scanned articles.
   • Portfolio — donut allocation chart, position manager, and AI suggestions scoped to your current book.
   • News — non-paywalled RSS headlines grouped by recency for each holding.

Yield/
├── Frontend/          # Chrome extension (TypeScript + esbuild)
│   ├── src/
│   └── manifest.json
│
├── app/               # React dashboard (Vite + TypeScript)
│   └── src/
│       ├── pages/      # Analysis · Portfolio · News
│       ├── components/ # SuggestionCard · Sidebar
│       └── utils/      # api · kelly · format
│
└── backend/           # FastAPI engine
    ├── main.py         # REST API + static file serving
    └── engine.py       # FinBERT · embeddings · theme/company detection

cd Yield/backend
pip install -r ../requirements.txt
uvicorn main:app --reload --port 8000

The engine warms up FinBERT and the sentence-transformer on first start (~10 s).

cd Yield/app
npm install
npm run dev          # → http://localhost:3000

Build and serve from FastAPI instead:

npm run build        # outputs to backend/static/
# then visit http://localhost:8000

1. cd Yield/Frontend && npm install && npm run build
2. Open chrome://extensions → Enable Developer mode → Load unpacked → select Yield/Frontend

Each card shown for an impacted holding contains four key pieces of information:

Half-Kelly (f* × 0.5) is used to account for model uncertainty. b = 1.5 reflects a typical equity reward-to-risk ratio. For informational purposes only — not financial advice.

The dashboard keeps the last 3 scanned articles in localStorage. The selector at the top of the Analysis page lets you flip between them to compare how different news events affect your holdings over time.