Dhruvin Meshiya

I’m an AI & Machine Learning Engineer with 3 years of hands-on experience building end-to-end systems that go from data ingestion to deployment. I focus on computer vision, LLM/Vision-LLM pipelines, real-time video analytics, and multi-modal AI workflows that are reliable in production.

At work, I’ve developed real-time object detection & tracking pipelines using YOLOv8/YOLO-NAS with ByteTrack / ByteSORT, improving tracking stability for video analytics. I’ve also built 3D hand pose estimation workflows and explored 3D Gaussian Splatting to push advanced computer-vision capabilities.

I build agentic, retrieval-augmented solutions that support real business use cases—like an Agentic ERP Automation & RAG chatbot with CrewAI and Langchain, multi-format ingestion, vector DBs, and an interactive speaking avatar for enterprise decision support. I implement OCR/Document Intelligence pipelines using Vision-LLM + prompt tuning to output structured JSON (e.g., receipt extraction with fields like totals and payment details) that teams can directly integrate.

I’m equally comfortable designing streaming architectures and production-style services, including low-latency multimodal agents using WebRTC with a dual-server FastAPI design, and end-to-end pipelines connecting STT → Vision-Language models → TTS. My approach is always modular, observable, and built to deliver fast, accurate outputs under real-world constraints.