Experience

Where I've worked

Sep 2025 - Present

AI/ML Engineer & Researcher

Particulate Systems Lab (PSL), Rutgers University

Building physics-informed machine learning systems and AI tools for pharmaceutical manufacturing research across simulation, optimization, NLP, and full-stack deployment.

  • ●Ported a legacy MATLAB granulation codebase into Python ML pipelines, cutting simulation runtime by 35% and validation time by 60%, with automated benchmarking that verifies numerical parity against the original models.
  • ●Trained Gradient Boosting and Random Forest surrogate models on 125 DEM simulation cases that replace multi-hour physics simulations with sub-second predictions of manufacturing outcomes, validated against held-out simulation runs.
  • ●Owned CI/CD, environment setup, and deployment for shared research tools across GitHub Codespaces, JupyterHub, and SLURM-based university HPC systems, used daily by NSF program managers and lab researchers.
  • ●Automated reviewer assignment with conflict-of-interest checks for NSF grant panels using PuLP optimization and SentenceTransformer + TF-IDF semantic matching, replacing a days-long manual process with an 8-app Streamlit platform.
  • ●Built an AI tool grouping and summarizing NSF proposals by topic using KeyBERT + fuzzy c-means clustering and a 4-model HuggingFace summarization pipeline (BART, Pegasus, T5), removing manual review across 30+ proposals per cycle.
  • ●Deployed NutriPulse, a production nutrition app with dual-API food search (CalorieNinjas + USDA, 380K+ foods), custom 5-dimensional relevance scoring reducing match errors by 75%, and Plotly analytics dashboards.
Oct 2025 - Nov 2025

Software Engineer (Contract)

ContinuumCare.ai

Built an AI-powered RAG-based health assistant using Python, LangGraph, and the Claude API.

  • ●Developed a RAG health assistant integrating 6+ medical guideline sources (AHA, SSATHI, NICE, PRANA) to generate evidence-based insights with 2-3 citations per response.
  • ●Raised response accuracy from 74% to 88% on a 150-question evaluation set through guideline-aware query rewriting over a 3-node LangGraph RAG pipeline (profile loading, guideline retrieval, response generation) with the Claude API, dynamically selecting guidelines by patient demographics and query intent to reduce hallucination risk, while cutting median response time from 2 minutes to 45 seconds with response streaming and result caching.
  • ●As the sole engineer, built and deployed a Streamlit web application in eight weeks supporting multi-profile management, structured health data uploads (labs, vitals, history), real-time AI health summarization, and PDF report generation, forming the MVP for the company's digital health assistant.
Sep 2023 - May 2025

Teaching Assistant

Dept. of Computer Science, Rutgers University

Delivered CS110/CS170 lab sessions and academic support for 200+ undergraduate students across 4 semesters of foundational computing.

  • ●Taught Scratch, HTML/CSS/JavaScript, Excel, and SQL queries through live coding demos and interactive labs, earning 95% positive feedback on end-of-term evaluations.
  • ●Rebuilt lab materials and assessments to be screen-reader accessible for a visually impaired student, restructuring visual code walkthroughs and diagram-based exercises into non-visual equivalents, contributing to strong course performance relative to cohort benchmarks.