Backend, data, and AI engineer who ships end-to-end systems — from data ingestion and ETL through API design, LLM orchestration, and automated evaluation. I like working where software has to make messy real-world information usable: city rules, market signals, exam prep, security data, APIs, and decision support. My strongest work combines backend engineering, data pipelines, and AI-assisted systems, with enough frontend/mobile skill to ship the full thing.

• Backend APIs, rule engines, and data-driven services
• Multi-agent AI systems, orchestration, and decision pipelines
• RAG and LLM-evaluation pipelines
• ML/data pipelines for classification, preprocessing, and evaluation
• Turning fuzzy requirements into usable, testable software

Runs seven specialist Claude agents concurrently every hour (technical, macro regime, sentiment, funding/whale, risk, news, breakout); an LLM orchestrator weighs their signals and places limit orders across BTC, ETH, SOL, and ZEC — unattended, with Telegram alerts on every order, fill, and close.

• Risk engine built to run live safely: ATR-based stops tuned per asset, drawdown circuit breakers that cut position size at −5% and halt all trading at −12%, plus funding-rate, velocity, and BTC-correlation vetoes.
• Validated honestly instead of curve-fitting: a full-year signal scan (371 trades) and walk-forward testing across three market regimes — which showed only one of four assets carries a real out-of-sample edge, and reshaped what the system trades.
• Writes to an Obsidian "second brain" nightly (trade notes, agent decision logs, backtest summaries), structured to become a RAG knowledge base for the orchestrator.
• Python · Claude API · Pandas · Coinbase Advanced Trade · Streamlit · Live Dashboard

Cross-platform Flutter app for the Architect Registration Examination, backed by an LLM quality pipeline.

• LLM-evaluation pipeline that audited the app's 1,100-question bank across five quality dimensions (judgment vs. recall, distractor plausibility, answer–explanation consistency, answer leakage, embedding-based duplicates) — full audit ran under $8 via the Anthropic Batch API and produced a ranked review worklist for a licensed architect.
• RAG generation loop: retrieve source text, generate a grounded question, pass it through the same five-grader gate with automatic distractor repair — ~80% of generated questions survive the gate at roughly $0.07 each, with a human reviewing samples before anything ships.
• App itself: 1,100-question bank across 7 divisions, SM-2 flashcards, mock exams, and an AI Coach (chat + voice) on Firebase Cloud Functions — server-side token quotas via atomic Firestore counters, App Check, keys in Secret Manager, GitHub Actions CI.
• Flutter · Firebase · Python · Claude API · sentence-transformers · pydantic

REST API that runs an ETL pipeline over NYC Open Data and evaluates 8+ curb-side parking-rule types, returning safe / caution / blocked decisions across 4 vehicle profiles.

• Modular rules engine, in-memory TTL caching, Swagger docs, 20+ unit tests, deployed live on Render.
• Python · FastAPI · NYC Open Data · Render

End-to-end ML pipeline (preprocessing, feature engineering, model training and evaluation) for vulnerability detection on real-world datasets, with measurable improvement over baseline. Recognized contributor on the MassLab research blog.

• Python · scikit-learn · Pandas · NumPy

B.S. Computer Science — Brooklyn College, CUNY (December 2025) Dean's List (Fall 2025). Coursework: Machine Learning, Database Systems, Operating Systems, Design & Implementation of Large-Scale Applications, Data Structures.

I'm focused on backend, data, and AI engineering roles where I can build reliable APIs, automate complex workflows, and ship systems that turn messy data into real decisions. I'm especially drawn to work that combines data pipelines, AI assistance, and developer-facing or decision-support systems.