AI Integration Lead and Machine Learning Engineer specializing in designing and deploying scalable AI systems across enterprise and research environments. Currently leading AI and automation integration at MPAC, modernizing property assessment workflows through LLMs, RAG pipelines, and large vision models. Expertise in blueprint understanding, document intelligence, semantic search, and real-time IoT data frameworks. Proven ability to translate research-grade AI into production-ready systems, aligning technical innovation with organizational goals.
GPA: 4.22/4.3 · Official Transcript
Thesis: SensorsConnect: World Wide Web for Internet of Things
Real-time IoT search engine powered by LLMs and RAG enabling natural language queries across heterogeneous sensor systems. Implemented a semantic search pipeline using Sentence-BERT and HNSW indexing, reducing query latency by 73%. Manages over 37,000 real-time IoT documents across 500+ service types in MongoDB with geo-indexing. Achieved 92% top-1 accuracy in complex intent detection, surpassing Gemini in benchmarks. Deployed on AWS using Docker Compose and Traefik with automated HTTPS and scalable reverse proxy management.
Retrieval-augmented generation (RAG) system combining vector search with LLMs to deliver context-aware movie recommendations, reducing hallucination rates by ~30%. Built a semantic search pipeline enabling retrievals from 1,000+ documents. Applied prompt chaining techniques to improve answer relevance, validated through precision metrics and user feedback. Built with OpenAI API, Pandas, and Tenacity for resilient real-time API usage.
End-to-end NLP pipeline using PyTorch and Hugging Face Transformers with LoRA adapters (PEFT) to fine-tune GPT-2 on the AG News dataset while keeping the base model frozen. Reduced training time and memory usage by over 60% vs. full fine-tuning. Boosted classification accuracy from 83.16% → 88.95%, demonstrating PEFT effectiveness with minimal compute.
Open-source Python package generating interactive 2D and 3D Radial Visualization (RadViz) plots for high-dimensional datasets using Plotly. Enables data scientists and analysts to discover hidden clusters, outliers, and trends through intuitive visual interfaces, improving model explainability and decision-making in analytics workflows.