Create your free profile

About

AI/ML Embedded Software Engineer specializing in the intersection of deep learning and hardware-aware inference. Proven experience building end-to-end MLOps pipelines, deploying models across NVIDIA Jetson, Hailo-8, and AMD Versal, and optimizing computer vision systems for real-world edge environments. Passionate about transforming heavy ML architectures into high-throughput, low-latency production deployments.

Experience2

  1. AI / ML Embedded Software Engineer

    Nov 2025 — Present

    Little Place Labs (LPL)

    Engineered automated conversion pipelines for PyTorch, TensorFlow, and ONNX models to optimized deployment formats including TensorRT and Hailo HEF, optimizing throughput and hardware efficiency. Built a model inspection platform and automated layer-testing agents to reduce retraining and speed feedback cycles. Architected an internal dataset intelligence platform using FastAPI, React, and Vite, improving data scientist productivity, recognized with 3rd place in company hackathon. Established MLOps infrastructure with Docker/Podman standards and integrated MLflow, ClearML, and model registries with documentation.

  2. AI & Computer Vision Intern

    Apr 2024 — May 2024

    PANTECH.AI

    Delivered 7 end-to-end ML projects including plant disease detection with 500+ classes and speech emotion recognition achieving 85% accuracy. Managed dataset preparation, model training, optimization, and deployment pipelines.

Projects3

  • Edge-AI Enabled UAS for Search and Rescue (SAR)

    Designed a two-drone swarm using Jetson Orin companion computers and Pixhawk controllers communicating via MAVLink/UART. Developed a custom Python GNC stack with dynamic path planning and Offboard Control. Deployed YOLOv8 for real-time human detection at 30 FPS with GPS geotagging and built a resilient WiFi/4G telemetry bridge for BVLOS communication.

    • Jetson Orin
    • Pixhawk 6C
    • YOLOv8
    • ROS2

  • Personalized AI Shadow – LLM Fine-Tuning

    Built a personalized LLM assistant replicating tone and reasoning style. Curated a large 3.4M entry technical and reasoning dataset (1.9GB). Fine-tuned the Mistral-7B model over 137+ hours achieving a loss of 0.35. Merged and quantized the model to GGUF format for efficient offline CPU deployment.

    • Mistral-7B
    • QLoRA
    • Hugging Face

  • VisionNet – Edge AI Drone Platform

    Developed an AI drone system for crack detection and solar panel inspection achieving 99.3% accuracy with 250 ms median latency. Applied TensorRT and INT8 quantization for real-time edge inference on Jetson Nano and Raspberry Pi hardware.

    • Jetson Nano
    • Raspberry Pi
    • TensorRT

Education1

  • B.Tech

    2025

    Techno India University Kolkata

    Computer Science (AI/ML)