ML engineer and researcher interested in LLM evaluation, AI Safety, personalization, and agentic systems.
M.S. in Artificial Intelligence & Innovation @ Carnegie Mellon University (May 2027)
- 🗄️ Agentic NL2SQL for Oracle Database (MSAII capstone with Oracle): schema-aware retrieval, dialect adaptation, and execution-based verification, evaluated on Spider 2.0, BIRD, and BEAVER.
- 💼 ML Engineer Intern, Klaviyo — Advanced ML Technology (Summer 2026). Built content-aware deep learning models that fold NLP embeddings of email content into multi-input architectures for send-time optimization (contextual bandits) and audience optimization; maintained the RL pipeline behind send-time optimization and contributed to a next-gen recommender system.
- 🛡️ LLM safety & unlearning. Developed SiMATU, a causal-tracing-based, layer-adaptive unlearning method for restoring safety alignment in compressed LLMs (Llama-2-7B-Chat; SafetyBench, AdvBench, MMLU, GSM8K).
- 🕸️ Euler, a graph-based multi-agent framework.
- 🎬 gradients-of-the-galaxy: a production-style movie recommender (LightFM, Kafka, Polars) with offline evaluation, blue-green deployment, and MLflow provenance tracking.
- A Pothole Can Be Seen with Two Eyes: An Ensemble Approach to Pothole Detection, Machine Vision and Applications (first author). link
Modeling: PyTorch · Hugging Face · RLHF / PPO / GRPO · contextual bandits · RecSys
LLM systems: RAG · LangChain · LLM-as-judge evaluation · multi-agent pipelines
Infra: Spark / EMR · MLflow · Kafka · Docker · HPC (Slurm on PSC Bridges-2)
Languages: Python · SQL · C++ · Java
