Taha A.

AI Engineer| Production RAG Systems & AI Agents

I ship AI to production — systems that run under real load, handle real data, and stay maintained after the contract ends. What I've built: MOKWN — built solo, from scratch, now live with 1,000+ electronics SKUs across Arduino boards, ICs, sensors, and power components. Includes Volt, an AI shopping assistant that handles semantic part search, BOM automation, and visual component identification — so a customer can find a part without knowing its exact name, or upload a photo of an unlabeled component and get a match. Stack: Qdrant, Azure OpenAI, React, Supabase. Tender & Proposal System — built for enterprise clients in the Saudi market who were manually reading and drafting responses to RFPs running to hundreds of pages, in Arabic and English. Standard PDF extraction fails on a lot of Arabic documents because the text layer is often broken or missing, so I built a vision-first pipeline that reads rendered page images instead, then runs structured extraction and retrieval-backed drafting. Every generated proposal goes through a human review step before it reaches the client — nothing ships unchecked. How I work: My focus is RAG pipelines and AI agents — retrieval architecture, chunking strategy, embedding selection, reranking, HITL integration, and the FastAPI/async backend infrastructure that ties it all together. I know where these systems break: context contamination, placeholder leakage, hallucination under long-context retrieval, chunk boundary failures. Catching those before a client does is most of the job. What I work with: LLMs & Embeddings: Azure OpenAI, Google Gemini/ADK, Cohere, Ollama, Hugging Face Vector Stores: Qdrant (primary), semantic search, hybrid retrieval Backend: Python, FastAPI, async architecture, structured logging Frontend: React, TypeScript — enough to own the full stack when needed Infra: Supabase, Azure, prompt versioning, RAG output auditing I work best on projects where the AI layer needs to actually function — not impress in a slide deck. If you're moving from proof-of-concept to something real, let's talk.