Applied work at the intersection of AI agents, systematic trading, and quantitative research
Numa — AI Stock Research Terminal
Python · FastAPI · Embedded LLM Analyst · FRED · SEC EDGAR · PWA
Full-stack financial research platform. FastAPI backend running nine parallel data modules per ticker — quotes, financials, technicals, options flow, SEC EDGAR insider filings, news sentiment, peers, earnings, analyst ratings — plus a FRED macroeconomic layer. Embedded LLM analyst receives a dynamically constructed full-context system prompt on every call, with per-call token-cost tracking. Deployed as an installable PWA.
9Parallel Data Modules
7Pre-built Analyses
Multi-Model AI Orchestration Framework
Python · Multi-LLM Orchestration · FastAPI · Azure · Pydantic
Reusable orchestration layer for dispatching agents across multiple LLM providers in parallel: per-agent timeouts and fallback handling, structured tool-calling against typed schemas enforcing validated outputs, and a deterministic consensus aggregator chosen for auditability over model-based voting. Adding a new agent or data source is a single adapter, not a rewrite. Powers Synapse's equity research pipeline and Numa's embedded analyst.
Vero — Multi-Agent Physical Therapy System
Next.js · TypeScript · Claude Sonnet · Claude Vision · ElevenLabs · Supabase
Six-agent clinical decision-support system for physical therapy, built in 12 hours with typed inter-agent contracts. Claude Vision with MediaPipe pose detection, voice guidance via ElevenLabs. 2nd place at the Caltech × UCLA × USC Claude Hackathon.
2ndof 50+ Teams
6Agents
12Hours to Build
Synapse — Multi-Model Autonomous Trading System
Python · Multi-LLM Orchestration · Alpaca · Docker · PostgreSQL
Autonomous equity research and trading system built on the orchestration framework above: parallel agent dispatch across 50+ US equities, real-time market data and regime signals feeding into typed schemas, and automated trade execution via Alpaca in a Dockerized pipeline with PostgreSQL state persistence and full execution logging.
50+US Equities
24/7Dockerized Uptime
Kalshi Prediction-Market Risk Engine
Python · Kalshi REST API · Azure · PostgreSQL · Risk Analytics
Live risk engine for a prediction-market trading platform: exposure caps, daily loss limits, confidence-adjusted sizing, and Azure-hosted monitoring dashboards. Edge model empirically calibrated on 72M+ historical contracts. Forensic post-mortem decomposing live-vs-paper PnL divergence into component risk factors.
72M+Contracts
2,490Resolved Trades
Python · LightGBM · Purged CV · Feature Engineering
Cross-sectional ML alpha signals on ~280M rows and 2.4k features for market-neutral 20-day return forecasts. Deployed live in Numerai's hedge fund with automated weekly retraining.
1.38Validation Sharpe
0.64Live Sharpe
300+Eras Tested
Macro Regime Duration Model
Python · statsmodels · Markov-switching · Nelson-Siegel VAR
3-state Markov-switching model detecting latent Recession/Growth/Expansion regimes in GDP, inflation, and unemployment. Integrated yield-curve forecast with regime-conditioned mean-variance optimizer.
1.08Sharpe
−45%vs Benchmark DD
87%Persistence
Delta-Neutral Options Market-Making Simulator
Python · Black-Scholes · LightGBM · cvxpy · Streamlit
Minute-level NBBO-aware delta-neutral simulator with BS theoretical pricing, Greeks calculation, and put-call parity residual tracking. Asymmetric spreads with inventory lean and delta-band hedging.
8–15%Target Fill Rate
550+Fills (800-min)
Python · Options Pricing · Real-time Data
Real-time arbitrage detection tool for put-call parity violations across options chains.
Post-Earnings Drift Backtest
Python · Event Study · Statistical Analysis
Systematic analysis of post-earnings announcement drift patterns across market cap segments.