Case 03 — SHORT RADAR: dual-lens crisis monitor
Two parallel crisis scores (0–100): a Burry lens for the US market built on 11 components, a Krall lens for Europe/Germany on 7 components. Live at shortradar.app as a 4-tier SaaS with signals, a sandbox portfolio and Telegram alerts.
The challenge
A short-hunter with 25 years of IT experience wanted his crisis thesis to be sellable — reproducible, with sources, weights and an audit trail. Not a single score, but two independent schools of thought running in parallel: the US vulnerabilities (Burry) and the European/German crisis (Krall). Requirements: FRED macro data and yfinance prices, weighted sector scorers, an ML layer (LSTM · XGBoost · HMM), a sandbox portfolio for risk-free testing, a Telegram bot, Stripe-billable tiers, a Next.js portal.
Architecture
A FastAPI core (:8002) with JWT auth and tier-based rate-limiting. Two parallel engines: the Burry engine (US, 11 weighted components across 4 sector scorers — Banks 60/40, CRE 55/45, Private Credit 50/50, High-Yield 35/65) and the Krall engine (EU/DE, 7 components — sovereign debt 22 %, euro banks 20 %, deindustrialisation 16 %, German real estate 14 %, stagflation 12 %, euro weakness 10 %, hard assets 6 %). ML layer: LSTM/GRU predictor, XGBoost short classifier, HMM 4-state regime detector, Isolation-Forest anomaly detector. Fetchers for FRED, yfinance, SEC EDGAR, options and news. Delivery: a Next.js 15 portal on dash.shortradar.app (:3030), a Telegram bot with 15 commands, Stripe for subscriptions.
Pipeline
Dual-lens score computation (every 15 min · full analysis every 4 h)
- 01Fetchers pull FRED (20+ series), yfinance (56 tickers), EDGAR (13F + Form 4), options and news — TTL cache (5 min / 6 h)
- 02Burry lens: 4 sector scorers (Banks 60/40, CRE 55/45, PC 50/50, HY 35/65) plus 7 direct FRED components → 11 weighted values
- 03Krall lens in parallel: 7 sector scorers for sovereign, euro banks, deindustrialisation, German real estate, stagflation, euro weakness, hard assets
- 04ML layer yields regime classification (HMM 4-state), anomaly score (Isolation Forest) and forward forecast (LSTM/GRU)
- 05Both aggregators weight into a 0–100 score each; a trigger matrix (9 rules, 4 h cooldown) sends tier-dependent Telegram alerts
Technology stack
Outcome
Live on shortradar.app (marketing + API) and dash.shortradar.app (portal + sandbox portfolio). Both scores reproducible, every sub-score auditable. 5 risk levels (CALM · MONITORING · ELEVATED · HIGH_RISK · CRISIS) delivered consistently over the REST API, a 15-command Telegram bot and a Next.js dashboard — Stripe controls delivery granularity through the 4 tiers (€0 · €149 · €499 · €1,999/mo). The previous oddini.bet domain 301-redirects to shortradar.app. Not financial advice — an analytical tool.