Mehroz M.

AI Agent Developer | AI Automation Agency | Voice AI Agents (Vapi, Retell, Twilio) | Claude Code | MCP Servers | RAG Pipelines | n8n, Make, Zapier | LangChain, LangGraph | OpenAI API, Claude API | Chatbot Development (WhatsApp, Slack, Web) | Appointment Booking & Lead Qualification Agents | CRM Automation (GoHighLevel, HubSpot, Zoho) | FastAPI, Python | Workflow Automation | API Integration Somewhere in your business there's a job an AI agent could be doing quietly in the background: answering the phone, sorting the inbox, keeping the catalogue current. I build those agents, and I build them to keep working long after the demo call. 33 contracts, 5.0 stars across 15 reviews, Top Rated, 100% Job Success. Ranked top 0.2% worldwide for LangChain and FastAPI on Upwork (UpworkMRR). Clients in healthcare, e-commerce, manufacturing, agencies, fintech and SaaS across the US, UK and EU. 𝗥𝗲𝗰𝗲𝗻𝘁 𝗽𝗿𝗼𝗷𝗲𝗰𝘁𝘀 → AI voice agent receptionist for a physiotherapy clinic. Bilingual, answers every call, books appointments straight into the practice diary, confirms by SMS. Live on a real number today. [Vapi, Claude, GoHighLevel] → Google Ads automation for a Shopify store. Reads real conversion data daily, adjusts bids and budgets on its own, refuses to raise spend without paid-conversion evidence. Running unattended since launch. [Python, Google Ads API, Shopify API] → Product catalogue pipeline for an industrial supplier. Web scraping, AI rewriting and publishing of 19,500 products and 1,500 collections with SEO titles, descriptions and alt text, six supplier sites migrated into one Shopify store. [Claude, Shopify API, Playwright] → LLM integration inside a legacy insurance CRM, where one wrong answer means real liability. Strict output validation, deterministic fallbacks, human sign-off on anything touching a policy. [OpenAI API, Python, CRM API] → Custom MCP server connecting Claude to a company's internal databases and tools. Data requests that used to wait in an engineering queue now get answered in plain English, with scoped permissions per tool and every call logged. [MCP, Claude, PostgreSQL] → WhatsApp automation for a UK manufacturer. Customers message at 2am, orders land in the right inbox, nobody on the team touches it. [WhatsApp Business API, Node] it. [WhatsApp Business API, Node] → Multi-user exchange platform with a matching engine, ledger and transaction history. Passed a 17-point acceptance run before launch. [FastAPI, PostgreSQL, Next.js] → AI content pipeline publishing three SEO-researched blog posts a day across two WordPress sites in two languages, 250+ published, every topic traced to a real search query. [Claude, WordPress API, Railway] → Hermes Agent and OpenClaw personal assistant setups connected to Gmail, Slack, WhatsApp and CRMs, hardened so they keep running after handover. 𝗪𝗵𝗮𝘁 𝗜 𝗯𝘂𝗶𝗹𝗱 𝘄𝗶𝘁𝗵 AI & Agents: Claude, Claude Code, Claude API, OpenAI API, GPT-4o, LangChain, LangGraph, MCP servers, multi-agent orchestration, prompt engineering Voice AI: Vapi, Retell AI, Twilio, OpenAI Realtime, ElevenLabs, Deepgram RAG & Document AI: Pinecone, pgvector, ChromaDB, hybrid search, PDF parsing, OCR, structured data extraction Automation: n8n, Make, Zapier, GoHighLevel, WhatsApp API, Slack, Airtable Reliability: retries, fallbacks, structured logging, eval sets, human approval gates Stack: Python, FastAPI, Node.js, PostgreSQL, Supabase, Docker, Railway, Vercel Integrations: Shopify, HubSpot, Zoho, Cliniko, Google Ads, Google Workspace, Stripe, Apify 𝗚𝗼𝗼𝗱 𝗳𝗶𝘁 𝗶𝗳 - You have a real process that runs today and want it automated properly - Budget is $1K+ for a build, or $400+/month for a managed agent - Someone on your side can make decisions and give access 𝗡𝗼𝘁 𝗮 𝗳𝗶𝘁 𝗶𝗳 - Price is your first filter - You want a demo, not a system 𝗪𝗵𝘆 𝗰𝗹𝗶𝗲𝗻𝘁𝘀 𝗿𝗲-𝗵𝗶𝗿𝗲 𝗺𝗲 → Every build ships with error handling, logging and approval gates. Nothing silently fails. → Replies in minutes, not days. → Code written for the next developer to read. → Most clients keep me on after handover to monitor and improve. A system that earns money deserves someone watching it. Send me the workflow, the failing repo or the integration that won't behave, and I'll reply with a short Loom on exactly how I'd fix it, before you spend anything.