Web Data UNLOCKED Hackathon · May 2026 · Track 2: Finance & Market Intelligence
An AI agent that synthesises pre-earnings signals from live web data — hiring trends, SEC filings, news sentiment, and financial metrics — into confidence-scored intelligence objects delivered through a live dashboard and REST API.
Public companies report earnings quarterly. In the days before that report, enormous amounts of information about what's likely in it are already publicly available — scattered across job boards, regulatory filings, and news articles. MarketPulse automates the entire process: find the signals, reason across them, and deliver a structured verdict before the earnings call happens.
Example output for META:
{
"ticker": "META",
"composite_signal": "BEARISH",
"confidence": 65,
"key_risks": [
"Rising AI investment expenses pressuring margins despite revenue beats",
"Market concerns about AI spending ROI timeline"
],
"recommended_action": "Monitor upcoming earnings call for specific AI monetization timelines.",
"reasoning": "Headlines about rising capex and investor concern outweigh the revenue beat signal."
}
React Dashboard → Flask REST API → Pipeline Orchestrator
│
┌─────────────────────┼──────────────────────┐
▼ ▼ ▼
Bright Data SEC EDGAR Yahoo Finance
SERP API (direct) (yfinance)
(news + hiring)
│
▼
Multi-Agent Claude Synthesis
(4 specialists + orchestrator)
│
▼
IntelligenceObject → ChromaDB → Cache → API
Four data sources feed four signal types. A multi-agent Claude system (4 specialist analysts + orchestrator) reasons across all signals simultaneously and resolves conflicts. Every result is stored in a vector database for trend detection across runs.
See docs/ARCHITECTURE.md for the full technical reference.
- Python 3.11+
- Node.js 18+
- Bright Data account with SERP API zone active
- CMU AI Gateway access (or direct Anthropic API key)
git clone https://github.com/YOUR_USERNAME/marketpulse-ai
cd marketpulse-ai
cp .env.example .env
# Edit .env — fill in Bright Data credentials and AI gateway keycd backend
pip install -r requirements.txt
python api/server.pyThe API starts at http://localhost:5000. On startup it pre-loads all 5 seed companies in the background.
cd frontend
npm install
npm run devDashboard at http://localhost:5173. The first account you register becomes superadmin.
| Source | Tool | What it provides |
|---|---|---|
| Google News | Bright Data SERP API | 10 articles per ticker — news sentiment + hiring signals |
| SEC EDGAR | Direct HTTP (no proxy) | Filing index text — management tone |
| Yahoo Finance | yfinance library |
P/E, market cap, revenue growth, profit margins |
Raw data
└── SERP articles → analyse_news() → NewsSentimentSignal
└── SERP hiring → analyse_hiring_from_news() → HiringTrendSignal
└── SEC text → [context passed to Claude]
└── Yahoo Finance → [context passed to Claude]
│
▼
Multi-agent synthesis:
news_agent + hiring_agent + filing_agent + financial_agent (parallel)
│
orchestrator → confidence + composite_signal + key_risks + recommended_action
│
▼
IntelligenceObject (Pydantic-validated, versioned, cached)
- Register at
/register— first account becomes superadmin, all others get user role - Login at
/login— JWT token issued, stored in browser - Admin dashboard at
/admin— user management, role promotion, live cache status - All intelligence API endpoints require a valid JWT token
| Upgrade | Status | Description |
|---|---|---|
| APScheduler | ✅ Active | Stale cache auto-refreshes every 6 hours |
| Persistent cache | ✅ Active | Cache survives server restarts via logs/cache.json |
| ChromaDB RAG | ✅ Active | Historical intelligence stored for trend detection |
| Multi-agent synthesis | ✅ Active | 4 specialist agents + orchestrator in parallel |
| Parallel fetching | ✅ Active | All 4 scrapers + all 5 companies run simultaneously |
| Rate limiting | ✅ Active | 300 req/day, 60 req/hour per IP |
| JWT authentication | ✅ Active | All endpoints protected, role-based access |
marketpulse-ai/
├── backend/
│ ├── api/
│ │ ├── server.py # Flask REST API — 6 endpoints
│ │ └── auth.py # JWT auth + SQLite user management
│ ├── pipeline/
│ │ ├── run.py # Orchestrator
│ │ ├── schema.py # Pydantic models
│ │ ├── scrapers/
│ │ │ ├── serp.py # Bright Data SERP API (news + hiring)
│ │ │ └── unlocker.py # SEC EDGAR + Yahoo Finance
│ │ ├── signals/
│ │ │ ├── news.py # Sentiment scoring
│ │ │ └── hiring.py # Hiring signal from news
│ │ └── synthesis/
│ │ ├── agent.py # Single-agent Claude + ChromaDB RAG
│ │ └── multi_agent.py # 4 specialists + orchestrator
│ ├── scripts/
│ │ ├── run_pipeline.py # Run all companies in parallel
│ │ └── backtest.py # Accuracy test against known outcomes
│ ├── tests/ # Component and end-to-end tests
│ └── utils/ # Logging, validation
├── frontend/
│ └── src/
│ ├── App.jsx # Routes + Dashboard
│ ├── api/client.js # Fetch helper with JWT headers
│ ├── context/AuthContext.jsx
│ ├── components/ # CompanyCard, SearchBar, ProtectedRoute
│ └── pages/ # Login, Register, AdminDashboard
├── docs/
│ └── ARCHITECTURE.md # Full technical reference
├── .env.example
└── .gitignore
# Single company pipeline
python -m pipeline.run --ticker AAPL --company "Apple Inc."
# All 5 seed companies in parallel
python scripts/run_pipeline.py
# Backtest against 10 known earnings outcomes
python scripts/backtest.py
# Tests
pytest tests/test_signals.py -v # no API calls needed
python tests/test_serp.py # live SERP test
python tests/test_unlocker.py # live SEC + Yahoo Finance
python tests/test_pipeline_all.py # full end-to-endHedge funds pay $50,000–$1,000,000/year for alternative data subscriptions that provide signals similar to what this pipeline produces. The infrastructure already existed — Bright Data handles the hard parts of web data access. What's new is the intelligence layer: a multi-agent Claude system that reasons across conflicting signals and produces a calibrated confidence score, not just a data dump.
Built for the Bright Data Web Data UNLOCKED Hackathon — Track 2: Finance & Market Intelligence.
Bright Data products used: SERP API · Web Unlocker · Web Scraper API
AI: Claude via CMU AI Gateway (OpenAI-compatible)
Stack: Python · Flask · React · Vite · Tailwind · SQLite · ChromaDB