Software Engineer at Techasoft Pvt Ltd (2025-02 – Present)
Edge AI Vision & Industrial Inspection (Britannia, ITC, Mahindra)
- Architected scalable edge AI infrastructure for real-time industrial defect detection using PLC-integrated camera systems and RTX/Jetson GPU inference pipelines.
- Lowered model inference latency by 70% (300ms to 85ms) for edge deployment by converting PyTorch models to optimized TensorRT engines.
- Built a multi-camera vehicle inspection system for Mahindra & Mahindra, reducing inspection latency by 50% through continuous real-time detection pipelines resilient to occlusions.
- Improved automated biscuit defect detection accuracy by 99.5% across 100+ production machines through YOLO fine-tuning with Optuna.
- Increased crowded-scene person tracking accuracy from 60% to 95% by tuning Kalman filters and stabilizing tracking constraints.
- Led a team of 8 engineers to build a microservices-based restaurant analytics platform processing daily sales data from multiple POS systems.
- Implemented AI-powered expense classification for automated bill processing by integrating OCR with Agentic AI workflows.
- Cut operational cloud expenses by 50% by streamlining ETL flows and removing unused datasets across AWS and Domo environments.
Software Engineer Intern at EdgeNRoots (2024-04 – 2024-09)
Custom OCR & Table Detection and Safety & PPE Inspection (PAC AI)
- Boosted OCR accuracy from 89% to 99.5% on niche industrial fonts by designing a custom feature engineering pipeline.
- Reduced inference latency from 100ms to 2ms by replacing heavy deep learning models with a lightweight morphological segmentation pipeline.
- Attained 96% detection accuracy for PPE compliance at airport restaurants by deploying a YOLOv7 and DeepSort tracking pipeline.
- Automated anomaly reporting via Twilio integration and managed the end-to-end model hosting on Paperspace.
