Engineering Archive
Selected builds across retrieval-augmented generation, computer vision, and cloud MLOps. Hover over any project frame to preview its silent 3–5s ambient capture slot.
Integrates Qdrant vector search, RAPTOR recursive tree summarization, and Neo4j Cypher traversal behind an intent router and cross-encoder re-ranker, containerized on FastAPI and Next.js.
Hierarchical tree-summarized retrieval and Neo4j knowledge graph over 70+ vascular aging research papers and medical device manuals, evaluated across 50 domain-specific queries.
Combines YOLOv8 expiry localization (92.3% mAP@0.5), PaddleOCR medicine extraction (93% accuracy), and voice scheduling in an offline-first mobile architecture. Second Runner-Up, BharatGen 2025.
Fuses MPRNet image denoising/deblurring and CLAHE preprocessing with YOLO object detection and Depth Anything V2 monocular depth estimation for indoor obstacle guidance.
Production classification service containerized with Docker and orchestrated on Google Kubernetes Engine (GKE), featuring MLflow experiment tracking and Evidently AI data-drift monitoring.
Automated technical viva simulator built with Next.js, Flask, ChromaDB, and Ollama (Llama 3) with text-to-speech interaction to evaluate responses against course lecture material.
Real-time surgical transit dashboard tracking organ pH, temperature, humidity, and GPS from Arduino sensors to Firebase, alongside a Next.js vector-embedding publication search portal for ACT Lab.