Turaboy H.

AI Product Engineer | RAG, LLM Agents, Voice AI | Python, Vue, React

I build production AI systems that ship inside real products, not notebooks: RAG pipelines, LLM agents, and voice-AI training flows, wrapped in the backend, frontend, and deployment they need to actually run. Recent proof: Haulmeter (haulmeter.online), a mobile-first platform I built solo that turns GPS-verified truck detention time into automated claims and invoices. And LotScope, a land-intelligence platform whose citation-grounded RAG assistant answers zoning questions with sources enforced in code. WHAT I DELIVER → RAG systems on your data: ingestion, chunking, embeddings, vector search (pgvector), and schema-validated, citation-grounded answers enforced in code so the model can't invent sources → LLM integration & AI agents: Claude and OpenAI APIs, tool use / function calling, agentic workflows, MCP servers, evals and guardrails → Voice AI that's cheap to run: speech-to-text (Deepgram), TTS (Azure Speech), and a cached-audio + LLM-evaluation architecture that cuts per-session cost dramatically and keeps response latency under 1.3 seconds → ML that earns money: demand forecasting, dynamic pricing, and valuation models (XGBoost, scikit-learn), containerized and retrained on schedule with Docker/Kubernetes → The full product around the AI: FastAPI or .NET backends, Vue/React frontends, PostgreSQL/PostGIS, offline-first PWAs, auth, tests, CI/CD RECENT PROJECTS → Haulmeter (founder, solo build): detention-recovery platform for US truck drivers. GPS-verified arrival/departure tracking feeds automated claim generation, invoicing, and receivables follow-up. Offline-first PWA (client-side image processing in web workers, real-browser Playwright coverage), modular FastAPI backend with feature-flag gating and an operator admin console with moderation and funnel analytics. Includes an English-proficiency voice-training module built on scripted scenario trees with pre-synthesized audio and LLM response evaluation. → LotScope (founder, solo build): vacant-land intelligence platform. Async FastAPI, PostgreSQL + PostGIS + pgvector, Redis workers, Vue 3/Quasar frontend. Its "Ask the Parcel" RAG assistant answers zoning and feasibility questions with citations enforced by schema validation. → Airline revenue management (Japan): XGBoost models forecasting bookings, load factor, and ADR across thousands of daily flights; Kubernetes-scheduled retraining; pricing recommendations built on competitor and event signals. → Fortescue Metals: led migration of an enterprise portfolio platform to Vue 3 + .NET 9 with Snowflake pipelines and earned-value tracking. WHY CLIENTS KEEP ME - 100% Job Success, Top Rated Plus, 1,300+ hours on Upwork - Full-stack engineer first: your AI feature arrives with a UI, tests, and deployment, not a demo script - Founder discipline: I run two production platforms of my own, so I think about unit economics, COGS, and validation — not just code - Enterprise habits from mining (Fortescue), government (PennDOT, King County), healthcare (a clinical app used in hospital emergency departments), and Fortune 500 (Kraft Heinz) work - I build with AI as well as build AI: my workflow runs on Claude Code with specialized review, testing, and security subagents, so I ship fast without cutting corners STACK Python (FastAPI, Django) · C#/.NET · Vue 3 & React · TypeScript · PostgreSQL/PostGIS · pgvector · Redis · Docker & Kubernetes · AWS & Azure · Claude API · OpenAI API · Deepgram · Azure Speech · XGBoost / scikit-learn Tell me what you're trying to build (or what's broken). I usually reply within a few hours with a concrete approach.