Dashboard of Feelow

Feelow 🦈

Cross-Market Intelligence Platform — Prediction Markets × Financial Sentiment

Feelow detects mispricings between what prediction markets (Polymarket) anticipate and what real financial markets (stocks, crypto) reflect. It fuses FinBERT sentiment, technical indicators, and Polymarket signals into a unified Market Mispricing Score displayed in a live Next.js dashboard.

📊 Features

Feature	Description
FinBERT Ensemble Sentiment	3-model voting (ProsusAI/finbert, DistilRoBERTa Financial, Sigma Financial SA)
Reddit Finance Sentiment	FinBERT scoring on reddit-finance dataset via HuggingFace
Real-Time RSS Ingestion	Yahoo Finance headlines per ticker
Technical Indicators	SMA, EMA, RSI, MACD, Bollinger Bands
Polymarket Agent Search	Gemini LLM searches Polymarket for prediction markets related to any company
Polymarket Scoring	Momentum, volatility, concentration, composite signal, cross-market correlation
Next.js Dashboard	Live KPIs, interactive price charts, news table with sentiment, Polymarket panel

🏗️ Architecture

feelow/
├── backend/                          # FastAPI unified API (port 8000)
│   ├── src/
│   │   ├── main.py                   # FastAPI app — all endpoints
│   │   ├── config.py                 # Central config (models, tickers, thresholds)
│   │   ├── full_pipeline.py          # Polymarket pipeline glue (agent-search → scoring)
│   │   ├── finance-data/             # Core financial modules
│   │   │   ├── sentiment_engine.py   # Multi-model FinBERT ensemble
│   │   │   ├── news_ingestor.py      # RSS headline fetching
│   │   │   ├── market_data.py        # yfinance price data loader
│   │   │   ├── technicals.py         # RSI, MACD, Bollinger, SMA, EMA
│   │   │   ├── gemini_agent.py       # Google Gemini search grounding agent
│   │   │   └── agent_orchestrator.py # Multi-step agentic pipeline orchestrator
│   │   ├── agent_search/             # Polymarket LLM search
│   │   │   ├── polymarket_pipeline.py
│   │   │   ├── orchestrator.py
│   │   │   └── scoring/              # Relevance, impact, novelty, sentiment, reliability
│   │   ├── polymarket-analysis/      # Advanced market scoring
│   │   │   └── market_scorer.py      # Momentum, volatility, concentration, composite signal
│   │   └── stock_analysis/           # Reddit-based FinBERT sentiment
│   │       └── api_finbert_transformer.py
│   └── tests/
└── webapp/
    └── UI-fr/                        # Next.js 15 dashboard (port 3000)
        ├── app/dashboard/page.tsx    # Main dashboard page
        ├── lib/ticker-context.tsx    # Global ticker state + API calls
        └── components/
            ├── section-cards.tsx           # KPI cards (price, sentiment, RSI, signal)
            ├── chart-area-interactive.tsx  # OHLCV price chart + Polymarket panel
            ├── data-table.tsx              # News headlines with sentiment badges
            └── app-sidebar.tsx             # Ticker selector (Tech / Finance / Crypto)

🔄 Webapp Data Pipeline

When the user selects a ticker in the sidebar, the Next.js frontend triggers parallel fetches to four backend endpoints:

User selects ticker
        │
        ├─→ GET /api/kpis?ticker=X
        │       MarketDataLoader → yfinance price + 7d change
        │       NewsIngestor → RSS headlines
        │       MultiModelSentimentEngine → avg_sentiment, signal
        │       TechnicalIndicators → RSI
        │       → SectionCards: price, Δ%, news volume, sentiment score, RSI, signal
        │
        ├─→ GET /api/price-history?ticker=X
        │       MarketDataLoader → OHLCV (yfinance)
        │       TechnicalIndicators → SMA20, SMA50, RSI, MACD, Bollinger
        │       → ChartAreaInteractive: area chart with period selector (7d/1mo/3mo/1y)
        │
        ├─→ GET /api/news?ticker=X
        │       NewsIngestor → RSS headlines
        │       MultiModelSentimentEngine → per-headline label + confidence
        │       → DataTable: sortable news feed with sentiment badges
        │
        └─→ GET /api/polymarket?ticker=X
                ticker → company name mapping
                PolymarketPipeline (Gemini) → relevant markets (pertinence 0–100)
                market_scorer → momentum, volatility, concentration, composite signal
                → ChartAreaInteractive: Polymarket panel with top markets + global score

SectionCards also fires a separate call for community sentiment:

POST /api/sentiment/score  { company }
        stock_analysis (Reddit FinBERT) → community sentiment score
        → displayed as "Reddit Sentiment" gauge card

🚀 Quick Start

Prerequisites

Python 3.10+
Node.js 18+
GEMINI_API_KEY (required)

1. Backend (FastAPI)

cd backend/src
pip install -r requirements.txt

echo "GEMINI_API_KEY=your_key" > ../.env

uvicorn main:app --host 0.0.0.0 --port 8000 --reload

API docs: http://localhost:8000/docs

2. Next.js Dashboard

cd webapp/UI-fr
npm install
npm run dev

Open http://localhost:3000.

📡 API Endpoints (used by the dashboard)

Method	Endpoint	Description
`GET`	`/api/kpis?ticker=X`	Price, pct_change, avg_sentiment, RSI, signal
`GET`	`/api/news?ticker=X`	Headlines with per-headline FinBERT sentiment
`GET`	`/api/price-history?ticker=X`	OHLCV + SMA, EMA, RSI, MACD, Bollinger
`GET`	`/api/polymarket?ticker=X`	Polymarket scored markets for the ticker
`POST`	`/api/sentiment/score`	Reddit-based FinBERT community sentiment score

🤖 Sentiment Models

Model	HuggingFace ID	Best For
FinBERT (ProsusAI)	`ProsusAI/finbert`	General financial sentiment
DistilRoBERTa Financial	`mrm8488/distilroberta-finetuned-financial-news-sentiment-analysis`	Financial news tone
Sigma Financial SA	`Sigma/financial-sentiment-analysis`	High-accuracy classification

🧪 Tests

cd backend
python -m pytest tests/ -v

Test file	What's covered
`test_score_polymarket.py`	Helpers, Market class, correlation, scoring pipeline
`test_full_pipeline.py`	Format bridge, pertinence normalisation, mocked flow
`test_backend.py`	HTTP endpoints, validation, full E2E

🏆 Hackathon Tracks

Best Use of Data (Susquehanna €7K) — Fuses RSS news, price data, and prediction markets into trading signals
Best Use of Gemini (€50K credits) — Gemini visual chart analysis + search grounding agent
Fintech Track (€1K) — Cross-market mispricing detection platform

Roadmap & Upcoming Features

🔄 Currently Implementing

Claude Agent — Deep reasoning with autonomous tool use (5 turns)
Deep Search — Advanced multi-source research
Live Social Media Monitoring — Real-time Twitter/X, Reddit, Stocktwits
Public REST API v1 — Consumable endpoints with API key authentication + rate limiting
WebSocket Streaming — Real-time push signals