Arijit De

Senior AI Engineer (PhD) with 6+ years building production-grade Agentic AI systems. Currently a Senior AI Engineer at mVizn, upgrading a computer vision system for Smart Port Project Management with multi-agent AI (LangChain, LangGraph, FastAPI) for intelligent routing, smart reporting and agentic workflow execution, increasing business efficiency by 70%. Fine-tuned BERT for intent classification and workflow routing (35% fewer tokens), deployed local LLMs with Ollama for secure workflows in critical port systems, and hardened agents with Pydantic tool-validation, guardrails and human-in-the-loop approval gates, cutting task failure rate by 60%.

Previously a Senior AI Engineer at Common Lab Ventures, building multi-agent solutions with multimodal RAG over 10M+ vectors of text, image and CAD data in Pinecone (92% recall@k, under 600 ms TTFT) using hybrid search and cross-encoder reranking. Designed a distributed PySpark/Databricks ETL pipeline that cut ingestion latency by 60%, led GCP deployments serving 3,000+ daily queries at 99% uptime, and added LangSmith and Ragas observability that lifted evaluation scores by 40%.

As an ML Engineer at mVizn, built an end-to-end 3D semantic segmentation system for LiDAR indoor point clouds (PyTorch, AWS CodeDeploy, EC2), reducing BIM creation time by 60%. Implemented FastAPI/asyncio backends with PostgreSQL (Amazon RDS) that increased concurrency by 40%, cut testing time by 40% with PyTest, and ran code reviews and mentoring sessions on CI/CD and deep learning.

Also worked as a Machine Learning Engineer at Synerjix on a Smart Moving Assistant for a USA-based client, using DinoV3 for object detection to estimate shifting costs directly from room photos. Built the backend with Python and FastAPI on a MySQL database, and deployed on AWS (SageMaker, EC2, EKS) with CI/CD and Docker, improving moving-cost estimation accuracy by 70% and stakeholder efficiency by 50%.

Awarded a PhD in Computer Vision and AI from the Electronics and Telecommunication Engineering Department of Jadavpur University (JU), Kolkata. Researched AI/ML-based detection, classification, segmentation and prediction algorithms using Deep Learning and Computer Vision for neurological disorders via radiological and histopathological imaging.

Alongside this research, worked as a Research Engineer at the Institute of Neurosciences, Kolkata, on AI-assisted neuroimaging — building custom 3D Computer Vision solutions for neurological disorder analysis with PyTorch and OpenCV (88.68% accuracy on clinical scans, 40% faster diagnosis), NumPy/Pandas ingestion pipelines, Matplotlib/Streamlit dashboards, and ONNX models deployed with Docker on AWS (SageMaker, EC2, EKS) that cut inference latency by 65%.

Earlier, as a Machine Learning Engineer at Mercedes-Benz R&D India (MBRDI), fine-tuned YOLOv3 for Vulnerable Road User detection (83% mAP50 in real time) and built a FastAPI/MySQL annotation-tracking dashboard backend, improving concurrency by 73%. Started as an Assistant Systems Engineer at Tata Consultancy Services (2014 – 2015), maintaining Java/MySQL backends serving 100,000+ users. Also has freelancing experience building and hosting a web platform for a Canadian client.

Deeply interested in Physical AI, Generative AI, Robotics and Computer Vision. Passionate about building AI-powered products and contributing to open-source projects. Always exploring new technologies and frameworks to enhance my skills and knowledge. Currently exploring Gemma, LeRobot and ROS2. Avid photographer, mythology enthusiast, casual gamer, and hardcore foodie.