YAHYA E.

AI Engineer who ships production AI: LLM apps, AI agents, RAG, voice AI, and the backend to run them. I take projects from idea to a deployed, monitored product, and I report results with numbers. WHAT I BUILD FOR CLIENTS - LLM applications & AI agents: tool-calling agents, MCP servers, chatbots, using commercial APIs or open-source models (LangChain, FastMCP) - RAG & search: document Q&A, knowledge bases, vector search (FAISS, embeddings, BM25), answer-quality evaluation - LLM fine-tuning: SFT, LoRA/QLoRA, DPO (TRL, PEFT), quantization (GGUF, AWQ) and self-hosting with vLLM to cut API costs - Voice AI: real-time voice agents, speech-to-text (Whisper, NeMo), text-to-speech, LiveKit/WebRTC - Computer vision & NLP: classification, object detection (YOLO), image captioning, content moderation - Backend & deployment: Python, FastAPI, Docker, PostgreSQL, Next.js, Vercel, Azure, CI/CD with PyTest - LLM evaluation & training data: benchmark design, test suites, preference data for RLHF PROOF - Built and operate a live voice AI SaaS end to end: real-time speech-to-speech pipeline, 31 agent tools across 6 MCP servers, authentication, usage metering and monitoring, with 0.9–1.7 s voice latency. Won 1st place in 3 competitions. - Fine-tuned a 6B tool-calling LLM that now serves production traffic; self-hosting it removed per-token API costs. - Trained a text-to-speech model ranked #2 of 11 open-weight systems on a public leaderboard, more intelligible than human recordings on its benchmark. - At Turing: designed coding benchmarks and training data for a frontier LLM. - On Upwork: delivered 4 fixed-price ML and API projects. - 25+ open models on Hugging Face (18,800+ downloads), 20 Docker images (16,000+ pulls), 2 sole-author arXiv papers. HOW I WORK Clear scope and milestones up front, regular updates, clean reproducible code, and a short handover report. Available 30+ hrs/week, Cairo time (UTC+3), with good overlap with Europe and US mornings. GitHub @NightPrinceY · Hugging Face @NightPrince · Live demos in my Portfolio below.