Smart money tracker + quant research terminal + defensive portfolio engine. Follows 20+ top hedge funds via SEC filings. 5-factor quant engine (Kalman filter, Markov chains, Hurst exponent) scoring 150+ stocks across the full AI infrastructure value chain. Beta-adjusted alpha + downside-protected portfolio construction (AQR Total Portfolio Approach) — win when the market is down. Congressional & insider trade monitoring. Real-time macro shock alerts. Delivered to your inbox every morning at 7 AM.

Built for investors who want to think like the world's best allocators: buy great businesses at fair prices and hold. Not a trading tool. No day-trading noise. Buffett/Klarman-style fundamental analysis enriched with quant scoring and AI.

"The stock market is a device for transferring money from the impatient to the patient." — Warren Buffett

The full picks-and-shovels stack behind the AI buildout — ~50 names mapped across every layer, from the chip to the power grid. This is where the next 10x ideas live before the market finds them.

Opportunity Score (0–100) — built to find asymmetric setups before re-rating:

Opportunity = Valuation (0–30)   ← GARP + forward P/E
            + Upside    (0–25)   ← analyst consensus target
            + Growth    (0–25)   ← revenue + earnings growth
            + Quant     (0–20)   ← 5-factor quant confirmation
            − Extended  penalty  ← demotes names up >80% on rich multiples

Labels: 🚀 EMERGING WINNER · 💎 UNDERVALUED · 🔥 HIGH UPSIDE · ⚡ MOMENTUM · ⚠️ EXTENDED Asymmetric "next Micron" setups are flagged ⭐ — high growth, fair value, real upside, not overextended.

python3 main.py --mode aii

A beta-adjusted alpha engine + downside-protected portfolio constructor, built on AQR's Total Portfolio Approach research. The goal: keep most of the upside, lose far less when the market falls.

Beta-adjusted alpha — every name's return is split into the free market ride (beta) and genuine unique edge (alpha), then scored by the appraisal ratio (alpha ÷ idiosyncratic vol — AQR's prescribed weighting metric):

Three-sleeve construction:

• 🟢 Alpha — high beta-adjusted-alpha growth (the return engine)
• 🔵 Defensive — low-beta ballast (BRK-B, COST, XLU, XLP, XLV, SCHD)
• 🟣 Convexity — the AQR crisis-diversifier playbook: DBMF/KMLM (managed-futures trend-following — AQR's #1 convex diversifier), BTAL (anti-beta), GLD, TLT

Weights ∝ appraisal ratio with a low-beta tilt (betting-against-beta), then the convexity sleeve is sized to pull whole-portfolio beta down to a target (default 0.60).

⚖️ Second method — Risk Parity (Bridgewater "All Weather"): the same universe is also built as an Equal Risk Contribution book, where position sizing ignores expected return entirely and instead equalizes each name's share of portfolio volatility (Maillard-Roncalli-Teiletche 2010, solved by cyclical coordinate descent). A 60/40 book is ~90% equity risk despite looking balanced by dollars; ERC scales low-vol ballast (bonds, gold, defensives) up and high-vol growth down until every holding carries its weight. The terminal renders both books head-to-head (annual return, vol, Sharpe, max drawdown, downside capture, beta) so you can see AQR's return-quality sizing vs Bridgewater's risk-balanced sizing on identical inputs.

Stress test — measures "win when the market is down": downside/upside capture, capture ratio, win-rate in down months, max drawdown vs SPY, and cumulative return through historical selloffs (2022 inflation bear, 2025 tariff shock).

python3 main.py --mode portfolio                  # target beta 0.60
python3 main.py --mode portfolio --target-beta 0.4  # more defensive

📘 Full concept primer: docs/PORTFOLIO_METHODOLOGY.md

Track what the world's best investors are actually buying — parsed live from SEC filings, congressional disclosures, and social signals. 23 funds, 7 tabs.

Funds tracked — 23 managers across 4 categories:

Every holding enriched with:

• 5-factor quant score → follow signal: ⭐⭐⭐ STRONG FOLLOW to ❌ AVOID
• QoQ change detection: NEW BUY / INCREASED / DECREASED / CLOSED
• Novelty scoring: non-consensus picks from Scion/Situational Awareness boosted; mega-cap consensus down-ranked

python3 main.py --mode whales

• 5-module quant scoring (each 0–100): Technical · Statistical · ML Trend · Risk · Fundamental
• 100+ stocks across 10 sectors · 75 ETFs (Tech, Bonds, Commodities, International, Factor, Dividend)
• 30 Hidden Gems: quantum computing, space tech, nuclear, biotech, eVTOL — small/mid-cap, under analyst coverage

A separate entry-timing overlay that combines momentum direction (MACD), momentum persistence (Hurst), and over-extension (Bollinger %) into one call — without touching the long-term quant score. The key is the pairing: MACD-up + Hurst-trending is a real trend worth entering, while MACD-up + Hurst-mean-reverting is fake momentum that will snap back.

Surfaced as BB% + Entry columns in the sector scan and screener tables, and in every stock's quant thesis.

LLM scans live RSS from Reuters, FT, WSJ, Bloomberg — identifies macro shock events (rate decisions, earnings, geopolitics) and surfaces the immediate trade setup for each.

GitHub Trending · TechCrunch · VentureBeat · arXiv · Reddit AI communities · YC companies → AI-generated bilingual briefing: top startups, research breakthroughs, funding intelligence, non-obvious trends.

Runs the full pipeline and delivers a dark-themed HTML report to your inbox every morning at 7 AM. Includes all sections: AI infra leaderboard, whale tracker summary, market intelligence, macro alerts, and tech briefing — in English and Chinese.

Sign up at console.groq.com — free tier, very fast, uses Llama 3.3 70B.

pip install -r requirements.txt

cp .env.example .env
# Required: GROQ_API_KEY
# Optional: GMAIL_ADDRESS + GMAIL_APP_PASSWORD for daily email

# AI Infrastructure value chain — hunt the "next Micron"
python3 main.py --mode aii

# Defensive Alpha — beta-adjusted alpha + downside-protected portfolio
python3 main.py --mode portfolio

# Smart money tracker — 23 hedge funds, 7 tabs
python3 main.py --mode whales

# Full run: AI infra → whales → sectors → tech briefing
python3 main.py --mode full

# Email mode: full run + deliver to inbox
python3 main.py --mode email

# Just stocks (sectors + ETFs + hidden gems)
python3 main.py --mode stocks

# Hidden gems only (~1 min, fastest)
python3 main.py --mode gems

# Tech briefing only
python3 main.py --mode briefing

python3 setup_cron.py
# Installs a cron job: every day at 7:00 AM (America/Los_Angeles)
# Sends the full report to the Gmail address in your .env

Every mode uses a built-in scrollable pager (less).

All data fetches run on a concurrent thread pool — stocks, ETFs, and whale holdings are fetched in parallel, not sequentially. SEC EDGAR requests go through a global rate limiter (8 req/s, under EDGAR's 10 req/s fair-access ceiling).

signalforge/
├── main.py                     # Entry point — 7 modes
├── config/
│   ├── universe.py             # AI infra value chain + 10 sectors + 75 ETFs + 30 hidden gems
│   ├── whales.py               # 23-fund registry with CIKs, styles, known-for
│   └── settings.py             # Screener thresholds, signal weights
├── stocks/
│   ├── ai_infrastructure.py    # AI infra value-chain scanner + Opportunity Score engine
│   ├── parallel.py             # Thread-pool helper — concurrent ticker fetches
│   ├── quant.py                # Master quant engine (5 sub-scores, 0–100)
│   ├── technical.py            # MACD, Stochastic, ATR, OBV, Fibonacci, ADX
│   ├── risk_metrics.py         # Sharpe, Sortino, VaR, CVaR, Beta, ML trend
│   ├── sector_scan.py          # Sector scanner + ETF table
│   ├── hidden_gems.py          # Small/mid-cap experimental picks
│   ├── screener.py             # Discount buy signals + GARP value picks
│   ├── backtest.py             # Walk-forward backtester + OLS weight optimization
│   └── data_enrichment.py      # Finviz scraper — short interest, analyst ratings
├── portfolio/                  # Defensive Alpha engine (AQR Total Portfolio Approach)
│   ├── factor_model.py         # Beta-adjusted alpha: CAPM beta, Dimson beta, appraisal ratio, convexity
│   ├── construction.py         # Appraisal-weighted + low-beta-tilt defensive constructor
│   ├── risk_parity.py          # Equal Risk Contribution constructor (Bridgewater All Weather)
│   ├── stress.py               # Downside/upside capture, drawdown, stress-window backtests
│   ├── data.py                 # Parallel multi-year price fetcher
│   ├── engine.py               # Orchestrator: data → factor model → construct → stress
│   └── display.py              # Rich tables + concept primer
├── whales/
│   ├── whale_display.py        # Whale tracker — 7 tabs, quant enrichment, novelty ranking
│   ├── sec_13f.py              # SEC EDGAR 13F parser — rate-limited, parallel, no re-downloads
│   └── social_signals.py       # Congressional trades, insider buying, StockTwits, news
├── scrapers/
│   ├── github_trending.py      # GitHub Trending
│   ├── feeds.py                # TechCrunch, VentureBeat, arXiv RSS
│   ├── reddit_ai.py            # Reddit AI communities
│   ├── yc.py                   # YC company directory
│   └── macro_news.py           # Reuters, FT, WSJ macro headlines
├── analysis/
│   ├── ai_analyst.py           # Sell-side style bilingual AI reports (EN + 中文)
│   ├── llm_client.py           # Multi-provider LLM client — Groq, Claude, Gemini, Ollama
│   ├── headline_trades.py      # Macro shock → trade idea LLM analysis
│   └── emailer.py              # Dark-themed HTML email via Gmail SMTP
├── run_daily.sh                # Cron entry point
├── setup_cron.py               # One-command cron installer (7 AM daily)
└── output/                     # Daily reports saved as Markdown

LLM_PROVIDER=groq
GROQ_API_KEY=your_key_here

# Optional — for daily email delivery
GMAIL_ADDRESS=you@gmail.com
GMAIL_APP_PASSWORD=your_16_char_app_password

This project demonstrates end-to-end financial engineering across four pillars:

🛡️ Defensive Alpha Portfolio Engine — CAPM factor model computing beta-adjusted alpha (appraisal ratio = alpha ÷ idiosyncratic vol), Dimson/lagged beta for true market sensitivity, betting-against-beta tilt, and a convexity sleeve (trend-following, anti-beta, gold, long bonds) sized to a target portfolio beta. Includes a second construction method — Equal Risk Contribution risk parity (Bridgewater All Weather, solved by cyclical coordinate descent) — rendered head-to-head against the AQR book on the same universe. Stress-tested with downside/upside capture, win-rate in down months, max drawdown vs SPY, and 2022 / 2025 historical selloff windows. Grounded in AQR's Total Portfolio Approach and Maillard-Roncalli risk-budgeting research.

🛰️ AI Infrastructure Value Chain — ~50 picks-and-shovels names scored across 8 layers (compute silicon, foundry, memory, networking/optics, servers/cooling, neoclouds, power, hyperscalers) with a custom Opportunity Score rewarding fair valuation + high growth + analyst upside before market re-rating.

🐋 Smart Money Tracker — Live SEC EDGAR 13F XML parsing for 20+ funds (Berkshire, Bridgewater, Renaissance, Tiger Global) with QoQ change detection, congressional trade monitoring (STOCK Act), and insider buying signals. Concurrent pipeline with global rate limiting completes in <90 seconds.

📊 5-Factor Quant Engine — Kalman filter (fair value tracking), Markov chain regime detection (BULL/BEAR/SIDEWAYS), Hurst exponent (mean-reverting vs trending), and multi-factor model (Value/Growth/Quality/Momentum) scoring 150+ stocks. Walk-forward backtested with OLS factor weight optimization. LLM-generated bilingual (EN/中文) analysis delivered by automated daily email.

For educational and research purposes only. Not investment advice. All models have inherent limitations. Do your own due diligence.

MIT License — free to use, modify, and distribute.