Mustafa Y.

AI Automation Engineer | AI Agents, RAG, Claude, n8n, Python, FastAPI

10,000+ compliance documents audited by a RAG pipeline I built, review cycles cut in half. 95% of a SaaS team's repeat tickets now closed by an agent I built. Content prep for a media brand with millions of followers down from a week to six hours. I am an AI Automation Engineer building AI agents, RAG pipelines and LLM chatbots on Python, Claude API, OpenAI API, n8n, FastAPI and PostgreSQL. Top Rated Plus, 100% Job Success, $175,000+ earned, 4,500+ hours across 30 contracts. Most AI builds pass the demo and fail when real documents, real volume and real users arrive. Every build I ship comes with an evaluation harness, so you measure accuracy yourself instead of trusting me. ———————— WHAT I BUILD ———————— → AI agents and multi-agent workflows Tool calling, human-in-the-loop approvals, escalation rules, audit logs. Claude API, OpenAI API, FastAPI, PostgreSQL, Docker. → RAG and AI knowledge bases Document ingestion, hybrid search, reranking, citations, NLP classification. pgvector and hosted vector stores. → AI chatbots and support automation Email, Slack, WhatsApp Business API and web widgets, with intent classification and CRM or helpdesk integration. → Workflow automation n8n, scheduled agents, webhook and OAuth integrations with Google Workspace and CRMs. → AI features inside existing SaaS React, TypeScript and JavaScript front ends on Python back ends, deployed on AWS with CI/CD and monitoring. ———————— HOW I BUILD ———————— → Scope against your real data Not a sample, not a demo set. If your documents are messy, you find out in week one instead of week six. → Ship an evaluation harness with the build You get a test set and accuracy numbers you can run yourself. You measure whether it works instead of hoping it does. → Design for the failure case first Retrieval misses, model drift, rate limits, malformed inputs. The system degrades visibly rather than inventing an answer. → Hand over something you own Documented code, deployment you control, no dependency on me to keep it running. ———————— WHAT THIS HAS LOOKED LIKE ———————— → Carbon-credit compliance platform RAG pipeline on Claude API auditing project documents against multiple certification standards across 10,000+ projects. Review cycles cut roughly in half. Running since early 2025. → Media brand with an audience in the millions Multilingual content pipeline with voice-matched script generation and CTR-scored assets. Content preparation that used to take a week now takes six hours. Running since early 2025. → SaaS support team Email and Slack agent with intent classification, refund and subscription handling, human escalation on anything ambiguous. Around 95% of repeat tickets now resolve without a human, so what reaches the team is the work that actually needed a person. → Hospitality concierge on WhatsApp Business API Multi-turn intake, location capture and live-availability booking. ———————— HOW WE WORK TOGETHER ———————— I am in Istanbul (UTC+3). That means a full working overlap with the UK and Europe, and 09:00-13:00 US Eastern most days. Two clients have run continuously with me since February and March 2025 across that gap - eighteen months without it becoming a problem. Async by default: a written update and a short Loom at every milestone, so you can see progress without booking a call. ———————— HONEST SCOPING ———————— If your idea is not ready, or the budget does not match the build, you hear it in the first reply rather than after you have spent money. I turn down work that will not succeed. That is a large part of why the Job Success score is still 100%. ———————— STACK ———————— Backend - Python, FastAPI, PostgreSQL, Docker AI - Claude API, OpenAI API, RAG, NLP, prompt engineering, evaluation harnesses, vector search Automation - n8n, scheduled agents, webhooks, OAuth, API integrations Frontend - React, TypeScript, JavaScript Infra - AWS, CI/CD, monitoring Most engagements run fixed-scope with a defined deliverable and timeline. Hourly when the work is genuinely open-ended. ———————— NEXT STEP ———————— Send me an invite or a message with a repo link, a document, or three sentences about what you are building. Two ways to start: 1. Paid 30-minute Architecture Review. You leave with a written build plan, a realistic cost range, and what I would cut. 2. Fixed-price 3-day Proof on your real data, before committing to the full build.