Ubair N.

I'm a Senior AI Engineer who takes AI from idea to production: LLM agents, RAG, Document AI, computer vision and automation. Top Rated with 100% Job Success, and 3+ years building AI at Cowlar (YC W17), Melior ITS and Siparadigm. WHAT I'VE SHIPPED AI Agents & LLM Apps • Natural-language database engine: dual-LLM orchestration (GPT-4o-mini + Claude) with graph-based schema intelligence on FalkorDB • FiscalChat: MCP server on the US Treasury API so users query fiscal data in plain English (Azure OpenAI, FastAPI, Next.js) • ReAct agent on Telegram that verifies Stripe payments and books Google Calendar slots on its own RAG & Knowledge Assistants • Monteur AI (Melior ITS): private RAG over huge machine-manual collections, every answer grounded to the exact document, page and section, on a self-hosted open-source LLM • Tender Assist: agentic system that drafts new tender responses from past submissions • CogniVision: chat with video using RAG + CLIP retrieval and T5 summaries Document AI & OCR • ADONUT (Siparadigm): complex layout extraction for medical and financial documents, tables included, with audit trails on every field • Resume extraction with RAG + Donut, 3x faster than human annotators • Edge price-tag OCR across retail stores: YOLOv8 + TFLite with DocTR, TrOCR and PaddleOCR Computer Vision & Edge AI • Optical fibre sorting (Cowlar): 95% color accuracy, sorting cut from 3 minutes to 8 seconds, 500% higher throughput, live in a US facility • Real-time CCTV analytics: queue and cashier wait times, shoplifting detection • Smart Cart action recognition on Android NPU with a GPU cluster backend • Satellite rooftop segmentation, edge classification and depth prediction for a measurement SaaS AI Automation & Integrations • GoHighLevel lead engine (n8n + Gemini): spam filtered out, replies in under 2 minutes, Hot/Warm/Cold scoring • Freshdesk + Shopify support agent: 80% faster responses, with human approval for risky tickets • AI chatbot with Stripe payments and calendar booking, kept running by the same client for 2+ years • Permit data platform scraping 60+ US city portals with Playwright + LLM parsing Generative AI • Stable Diffusion, ComfyUI and ControlNet pipelines for virtual try-on, face editing and image workflows HOW I WORK • I start from your data and the result you need, then pick the simplest model that gets there • Built for production: FastAPI, Docker, CI/CD and monitoring on AWS, Azure or GCP • Clear, regular updates. "Clear Communicator" is my most frequent client endorsement TOOLS LLMs: OpenAI, Claude, Gemini, DeepSeek, open-source models via Hugging Face | Frameworks: LangChain, LlamaIndex, MCP | Vector DBs: Pinecone, Chroma, pgvector | CV & ML: PyTorch, TensorFlow, YOLO, OpenCV, Roboflow | Automation: n8n, Make, Zapier, GoHighLevel, HubSpot | Backend: Python, FastAPI, Next.js, PostgreSQL, MongoDB Tell me what you want to build or automate, and I'll reply with a practical plan and an honest view of what's possible.