Animesh Bhardwaj

Optimized YOLOv8 and ArcFace pipelines (Python, TensorFlow, PyTorch, OpenCV) for defense surveillance, increasing detection mAP from 0.75 to 0.83 and recognition accuracy from 85.7% to 92.4%.

Converted

PyTorch (.pth) models to TensorRT (.engine) format and applied INT8/FP16 quantization and CNN tuning, reducing inference latency by 76% (250 ms → 60 ms) and increasing Jetson Nano throughput from 5 FPS to 20 FPS. Containerized AI inference components (YOLOv8 detection, ArcFace recognition) with Docker and orchestrated deployment on Kubernetes, standardizing the edge-AI inference environment across NVIDIA Jetson hardware.

Artificial Intelligence & Computer Vision Intern at Indian Navy Incubation Centre for Artificial Intelligence (INICAI) (2025-11 – Present)

• Optimized YOLOv8 and ArcFace pipelines (Python, TensorFlow, PyTorch, OpenCV) for defense surveillance, increasing detection mAP from 0.75 to 0.83 and recognition accuracy from 85.7% to 92.4%.
• Converted PyTorch (.pth) models to TensorRT (.engine) format and applied INT8/FP16 quantization and CNN tuning, reducing inference latency by 76% (250 ms → 60 ms) and increasing Jetson Nano throughput from 5 FPS to 20 FPS.
• Containerized AI inference components (YOLOv8 detection, ArcFace recognition) with Docker and orchestrated deployment on Kubernetes, standardizing the edge-AI inference environment across NVIDIA Jetson hardware.
• Used Claude Code, Codex, Antigravity IDE, and ChatGPT daily to debug TensorRT/CUDA integration issues, refactor inference pipeline code, and rapidly navigate unfamiliar surveillance codebases, shortening development cycles on the deployment pipeline.
• Enhanced multi-object surveillance by increasing tracking success from 78% to 89%, benchmarking YOLOv8, ArcFace, and DeepSORT, and documenting deployment guidelines for defense AI systems.

Artificial Intelligence & Computer Vision Intern at Indian Navy Incubation Centre for Artificial Intelligence (INICAI) (2025-06 – 2025-08)

• Built a real-time personnel detection and tracking framework processing live video streams using Python and OpenCV, and implemented CNN-based face recognition models with optimized preprocessing to improve robustness under varying illumination.
• Integrated detection, tracking, and recognition modules into a unified AI/ML surveillance workflow for robotics applications, establishing the foundation later extended with TensorRT optimization, Docker/Kubernetes deployment, and edge inference on NVIDIA Jetson hardware.

Full Stack Web Development Intern at BrainStorm Achievers (2024-06 – 2024-08)

• Built 3 core Django apps — including student registration, attendance/performance tracking, and course management — supporting 100–200+ students across 2–3 course programs.
• Improved application maintainability by designing modular Django backend components, integrating frontend interfaces, and optimizing database-driven workflows using SQL for CRUD operations, filtering, aggregation, and joins across related data.
• Tested, debugged, and deployed production-ready features in collaboration with the development team, improving application reliability and user experience.

B.Tech in Computer Science Engineering (Robotics & Artificial Intelligence) – Manav Rachna University (2022 – 2026)