AI/ML Engineer · Software Engineer — MS Data Science, Stony Brook University '26
- I build AI and ML systems that actually run — retrieval and agent pipelines, models with calibrated uncertainty, and the data platforms underneath them.
- Backend engineer by training, so the models ship: FastAPI/Spring Boot services, Postgres and vector stores, Docker, CI, and integration tests around the whole thing.
- Shipped production enrollment APIs and database workflows powering ARBenefits portal processes for 184K+ covered employees, cutting enrollment processing time by 40%.
- MS in Data Science from Stony Brook University. Project work spans LLM/RAG systems, reinforcement learning, experimentation and forecasting, and cloud data pipelines.
ML / AI
Systems & Platform
Also comfortable with: RAG/LLM tooling (Qdrant, ChromaDB, Ollama, sentence-transformers), experimentation and causal analysis (power, bootstrap CIs, guardrail metrics), DuckDB/BigQuery analytics engineering, Pandas/NumPy workflows, Testcontainers-backed integration testing.
One flagship from each of the four areas I work in.
|
PaperTrail Research-paper assistant that connects methods and concepts across papers — search by ideas, not keywords. 🔗 Repo AI & LLM Systems · RAG, Knowledge Graphs |
Geometry Dash RL DQN vs PPO vs genetic algorithms, trained in a 4000×-real-time headless sim and evaluated in the real game. 🔗 Repo ML & Modeling · RL, PyTorch |
A/B Experiment Analysis Pre-registered power analysis through guardrails, segment lifts and a ship/hold decision — validated against known ground truth. 🔗 Repo Data Science · Experimentation, Inference |
SMB Growth Intelligence GA4 events → layered ETL → point-in-time feature store → propensity model → data-quality engine. Local DuckDB, portable to BigQuery. 🔗 Repo Data & Platform · Analytics Engineering |
| Area | Project | What it does |
|---|---|---|
| AI & LLM Systems | intelligent-doc-rag | Self-hosted RAG over your documents — Spring Boot, Qdrant, Ollama. Grounded answers with enforced citations and multi-tenant vector isolation. |
| AI & LLM Systems | forensic-llm-wiki-obsidian | Markdown-first forensic investigation wiki that compiles raw evidence into evolving hypotheses, contradictions and reports. |
| ML & Modeling | Robust Image Classification | MC-Dropout uncertainty with a "tri-gate" abstention policy — entropy and mutual information calibrated on SVHN for out-of-distribution rejection. |
| Data Science | forecasting-baselines | Leakage-safe walk-forward benchmark of naive/MA/Ridge/XGBoost/LSTM on real market data. Reports the honest result: the random walk is not beaten. |
| Data & Platform | devops-platform-lab | Terraform + Kubernetes + Helm + GitHub Actions CI/CD with Prometheus/Grafana observability on a local kind cluster. |
| Data & Platform | MarketPulse · Uber Analytics · Azure E2E | Streaming and batch pipelines across Kafka, GCP/BigQuery/Looker, and Azure Data Factory/Databricks/Synapse. |