Sai Naveena S.

🎯 Computer Vision, LLM & Edge AI Engineer | 6+ years | 20+ production AI systems shipped I build Computer Vision, LLM inference, RAG, and AI Agent systems that run in real factories, clinics, and edge devices, on NVIDIA GPUs I deploy and tune myself. YOLOv11, SAM3, vLLM, TensorRT, LangGraph, FastAPI, React. Not demos, not notebooks. Systems that run every day. ⚙️ MY LATEST BUILD A defect detection system for one of India's leading tyre manufacturers. 8 conveyor inspection stations, running fully on edge hardware, zero cloud dependency, low-latency inference on the factory floor. 🏭 WHAT I'VE SHIPPED ⚡ Self-Hosted LLM Inference & RAG Private LLMs served with vLLM on NVIDIA GPUs (A100, RTX A5000): continuous batching, KV-cache memory tuning, FP8 quantization, token streaming, hybrid RAG with citation checks, and GPU workers that scale to zero. Measured on one A100 80 GB: a 14B model at 1,297 tokens/sec with 64 requests at once, about 10x a single-stream server. No data leaves your servers. 🔍 Auto-Labeling Platform Desktop app that lets non-technical staff annotate datasets and train models on their own GPU. SAM3 + Grounding DINO for zero-shot labeling, RF-DETR and YOLOv11 for training. 100% offline. 📹 Real-Time Compliance Detection Video analytics that verifies physical processes like hygiene and cleaning routines. Motion classification plus a homography-based coverage engine, with live heatmaps and overlays. 🏫 Special Education Management Portal Secure, role-based platform that digitized therapy assessment workflows. React, Vite, Tailwind, Supabase, PWA. 🧠 WHAT I DO 👁️ Computer Vision — detection, segmentation, tracking, classification. YOLOv11, SAM3, Grounding DINO, MediaPipe, OpenCV. ⚡ LLM Inference — vLLM serving, FP8 / INT8 quantization, batching, latency and cost-per-token tuning on NVIDIA A100, RTX and Jetson GPUs. 🤖 AI Agents — multi-agent systems with LangChain, LangGraph, CrewAI, MCP, Pinecone RAG. ✨ Generative AI — LLM integration, LoRA fine-tuning, self-hosted inference, production chatbots. 💻 Full-Stack — React, Node.js, FastAPI, Python, PostgreSQL, Supabase, REST APIs, dashboards. 🎙️ Voice AI — real-time conversational agents with Pipecat and Twilio. 🥽 AR/VR — 7+ XR apps in Unity and Unreal, including patented AR healthcare and VR industrial training. 📡 Edge AI & IoT — TensorRT FP16 / INT8 and ONNX on NVIDIA Jetson and embedded hardware, no cloud required. 🛠️ STACK Python · PyTorch · TensorFlow · vLLM · TensorRT · ONNX Runtime · Hugging Face · CUDA · OpenCV · FastAPI · React · Node.js · LangChain · LangGraph · CrewAI · Pinecone · Qdrant · Docker · PostgreSQL · Supabase · YOLOv11 · RF-DETR · Grounding DINO · SAM3 · MediaPipe · Unity · Unreal · Linux · NVIDIA GPU deployment ✅ HOW I WORK I deliver the whole product, not just a model. Data pipeline, training, backend, UI, deployment, and support after go-live. Clear milestones, weekly updates, and no surprises. 🎓 CREDENTIALS - M.S., IIT Madras (CGPA 9.33) - 2 patents filed - 3 peer-reviewed publications (ASME 2023, Springer) - Unity Certified Developer - 3rd Prize, 18th GDC INCUBATE National Competition - 2nd Place, CII Industrial Innovation Awards 2021 📩 Tell me the problem you're trying to solve and I'll tell you straight whether AI is the right fix, what it will take, and how long. Happy to start with a paid pilot.