AI Engineer at VisionRD

Developed and deployed vFuseAct, a dual-head MemoryTCN model with temporal segmentation for real-time, framewise action segmentation on Raspberry Pi Zero 2W at 30 FPS. Deployed a Forgery Detection System using OCR+LLM at a Toyota plant, achieving 98% accuracy and a 70% reduction in manual verification. Engineered a production-grade RAG-based marketing tool for Sazgar Motors enabling real-time conversational analysis of consumer sentiment and an 80% reduction in manual analysis. Implemented the GriTS evaluation framework with Landing AI to benchmark table structure recognition across document models. Built a PCB Defect Detection System inspecting 10–15 million PCBs per year in real time. Led continuous R&D initiatives to integrate cutting-edge CV/NLP research into production workflows.

Design Engineer at Ministry of Defense

Design of military-grade Laser Actuated Target System for realistic training using ARM-Cortex-M3, LRF modules, and LiDAR-Lite v3HP, reducing training ammunition expenditure by $76,000 annually. Developed an offline RAG system for secure document intelligence using LangChain, LLaMA 3, and ChromaDB for air-gapped conversational search over SOPs and manuals. Designed cost-effective EV Cart with solar, integrated IoT and Battery Management System; implemented sensor fusion for real-time inventory tracking and automated logistics planning for 3000+ vehicles.