AI / ML Engineer · production LLM systems, applied ML, and the research behind them
I build end-to-end LLM applications, RAG pipelines, agentic workflows, and confidence-aware systems that know when not to answer, on a foundation of applied ML (forecasting, recommender systems, fairness-constrained modeling). I'm completing an M.S. in Applied AI at Stevens Institute of Technology, and my work is backed by peer-reviewed research, including a Best Paper award (ICICC 2026, Springer) and an IEEE Xplore publication.
I use AI coding assistants as part of my workflow, the way most engineers do now, but the architecture, engineering decisions, and trade-offs in every project below are my own.
| Project | What it is | Link |
|---|---|---|
| Selective RAG with Conformal Abstention | Retrieval-augmented generation that scores its own confidence and abstains on low-confidence queries (AUROC 0.8515, 4.27× reliable-coverage gain) | Code (publishing soon) |
| VAULTS | Full-stack, LLM-powered portfolio-intelligence app — Next.js · TypeScript · Anthropic API · Upstash Redis | Live demo |
| AI Command Center | LLM-powered coaching & briefing app with agentic workflows and retrieval over user data | Live demo |
| FairLens | Algorithmic-bias audit across 800k+ U.S. mortgage records (HMDA) with fairness metrics, shipped as a web app | Live |
| Music Recommendation Ensemble | 88-model stacking ensemble (PySpark ALS + LightGBM LambdaRank) on 12.4M ratings — Kaggle 0.930 | Leaderboard |
| Cost-Sensitive HAR | Six classifiers built from scratch in NumPy with a Chow reject option — 94% accuracy at 2.7% abstention | Code |
Several production apps are kept private for security; live demos are linked above and code is available on request.
- Reinforcement Learning-Based Energy Dispatch for AI-Enabled Surgical Units under Variable Renewable Supply — Best Paper Award, ICICC 2026 (Springer). RF/XGBoost/LSTM/GRU forecasting (R² 0.9345) coupled with an RL dispatch agent.
- Digital Twin for Battery-Powered Medical IoT Fleet — IEEE Xplore, AIMLA 2026. Physics- and ML-based battery SoH/RUL modeling. DOI
- Working papers: decision-theoretic abstention for Human Activity Recognition · algorithmic fairness in U.S. mortgage lending
Focus areas LLM application development · RAG · agentic workflows · embeddings & vector search · conformal / selective prediction · recommender systems · time-series forecasting · fairness-constrained ML
Building confidence-aware LLM systems and applied-ML research oriented toward high-stakes and sustainability-focused decision problems (energy, fairness, reliability).
Open to AI / ML Engineer roles.