AI/ML Engineer and Researcher · MS Statistics - Data Science, Rutgers University
Portfolio · LinkedIn · Email · Open to opportunities (open to relocation)
I build ML systems at the intersection of research and real-world impact. Currently a AI/ML Engineer and Researcher at the Particulate Systems Lab at Rutgers University, working on physics-informed ML, pharmaceutical manufacturing simulation, and AI tooling for NSF grant workflows. I care about systems that are not just accurate, but actually useful.
- Physics-informed ML pipelines that replace MATLAB granulation simulations with Python, cutting runtime by 35% and validation time by 60%
- Gradient boosting and random forest surrogate models that turn multi-hour physics simulations into sub-second predictions
- CI/CD and deployment ownership for shared research tools across GitHub, JupyterHub, and university HPC infrastructure
- AI tools for NSF proposal processing: SentenceTransformers + KeyBERT clustering, PuLP-based reviewer assignment with conflict-of-interest checks
- SourceMD: agentic RAG fact-checker that scores AI-generated medical answers against real clinical guidelines, deployed across Vercel, Render, and Neon serverless PostgreSQL, with a 70% backend memory cut from switching to ONNX-quantized FastEmbed embeddings
- AI DevOps Agent: autonomous agent powered by Gemini 3 Flash that monitors live Vercel deployments, detects build failures and security issues, and opens a GitHub PR with a fix, built in a single day at Zero to Agent, a Cerebral Valley x Vercel x Google DeepMind hackathon
- NutriPulse: production nutrition tracker with dual-API food search across 380K+ foods
Languages
AI & LLM Systems
-6D28D9?style=flat&logoColor=white){kind=icon}
Backend & Frontend
ML & Data
Visualization
Cloud & DevOps
| Project | What it does | Stack |
|---|---|---|
| SourceMD | Agentic RAG fact-checker: verifies claims against ~844 chunks from clinical guidelines (NICE, AHA), scores SUPPORTED / UNSUPPORTED / CONTRADICTED, and returns a 0-100 trust score with a corrected, source-backed rewrite. Cut backend memory 70% with ONNX-quantized FastEmbed embeddings; deployed across Vercel, Render, and Neon | LangGraph, FastAPI, ChromaDB, FastEmbed, Neon, React, TypeScript, Docker |
| AI DevOps Agent | Autonomous agent that monitors live Vercel deployments, detects build failures and security issues, and opens a GitHub PR with a fix. Built from whiteboard to demo in a single day at Zero to Agent, a Cerebral Valley x Vercel x Google DeepMind hackathon | Next.js, Vercel AI SDK, Gemini 3 Flash, Supabase, GitHub API, Slack |
| NutriPulse | Full-stack nutrition tracker with dual-API food search across 380K+ foods, custom 5-dimensional relevance scoring that cut match errors by 75%, and Plotly analytics dashboards | Python, Streamlit, Plotly, CalorieNinjas API, USDA FoodData Central |
| BioNER | Multi-task biomedical NER model identifying genes, diseases, and chemicals across 5 benchmark corpora, reaching an F1 score over 90% | PyTorch, BioBERT, HuggingFace Transformers |
| Image Caption Generator | CNN + Transformer captioning model (BLEU = 0.80) trained on 120K+ images, deployed as a live Streamlit app | InceptionV3, Transformers, Streamlit |
| Growth Mindset Study | Causal inference on 10K+ student records using 5 estimators, including Causal Forests and X-Learner | EconML, scikit-learn, R |
- Clustering and Visualization of Research Proposals Using Sentence Embeddings and Unsupervised Learning, ongoing (NSF-funded program)
- Data-Driven Acceleration of PBM-DEM Granulation Model via ML-Based Prediction, ongoing
- AWS Certified AI Practitioner (Feb 2025)
- Google Cloud, 30 Days of Cloud (Oct 2021)