Work

Building production agentic systems for supply-chain planning using LangGraph orchestration, FastAPI services, and Azure OpenAI deployments with tool-calling, MCP integrations, and guardrails.

Shipping agents (DocumentAssist, Forecast Assist, Supply Assist, MaaS Agent) with end-to-end evaluation pipelines — Langfuse traces, LLM-as-judge scoring, and structured evals that reduced hallucinations across production workloads.

Delivered agentic services to Intel and Micron on Azure AI Foundry, contributing to $200K+ paid engagements.

Built quantitative trading platform for Kalshi and Polymarket: Stoikov inventory-aware market-making, favorite-longshot bias model, online calibration tracking, and self-improving policy optimization.

Developed ERCOT energy intelligence system with 4CP peak prediction (95% accuracy) and autonomous multi-agent framework using Claude tool-calling; demonstrated $24.5K/week value capture.

Implemented temporal graph networks (PyTorch) for private equity portfolio contagion prediction with BFS propagation and Rust risk engine via PyO3.

Designed LLM evaluation platform with 6 domain-specific evaluators (LLM-as-judge, factual, coherence) scoring across energy, trading, and PE domains.

Processed 5+ million daily energy transactions for California's $50 billion market by building CAISO data integration system that enabled real-time trading affecting 30 million residents.

Engineered 99.99% uptime data synchronization matching CAISO's publication frequency to capture millions in daily price arbitrage opportunities for energy traders.

Accelerated settlement processing by 95% (hours to minutes) through database optimization of California's entire power grid data, directly impacting energy costs statewide.

Built AI SaaS platform transforming research papers into interactive podcasts using LLM orchestration (OpenAI, Claude, Groq), ElevenLabs TTS, and custom prompt templating system with 50+ engineered prompts.

Secured meetings with Springer Nature and Wiley to pitch AI-powered academic content transformation, demonstrating product-market fit with major publishing industry leaders.

Developed full-stack platform with FastAPI backend, prompt library architecture, and PDF parsing pipeline that automated complex academic content into engaging audio experiences.

Built biomedical AI data platform processing 1TB+ drug and protein data using S3 storage and entity extraction pipelines that structured DrugBank, FDA, and ChEMBL datasets into searchable noSQL databases.

Processed 1M+ PubMed research papers with NLP pipelines for cancer research, enabling discovery of hidden drug-protein interactions through AI-powered pattern recognition and interactive dashboards.

Achieved successful acquisition by Xyant Services after presenting AI platform capabilities to Johnson & Johnson and NIH stakeholders, contributing to NSF grant development.