Martin L.

AI Agent Engineer | Orchestration, Claude MCP, LangGraph, GPT

Most "AI developers" can wire ChatGPT to a database. I build agent systems that survive in production — orchestrators with retries, state management, observability, and cost control. 6+ years as a Software Engineer, now full-time System Engineer building AI-driven SaaS at Mplus (DE), available for select Upwork projects. I'm Martin — Top Rated, 100% JSS, 5★ across all closed contracts. ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ WHAT I ACTUALLY SHIP → Production AI agents that don't fall over Stateful agents with retries, tool routing, error recovery, observability, and cost control — not a demo that breaks on the second user. → Custom orchestration when frameworks aren't enough LangGraph and Claude MCP where they fit. Custom architectures when your workflow doesn't fit a framework. I design the agent system — I don't just bolt one together. → Claude & OpenAI integrations done seriously Anthropic Claude with Tools and MCP, OpenAI Assistants & Agents SDK, function calling, structured outputs, streaming, hybrid model routing for cost/latency trade-offs. → Full-stack delivery around the agent Python for the AI layer, Node/TypeScript for APIs and web. Postgres, pgvector, queues, webhooks. The agent isn't the product — the system around it is. I build both. ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ CURRENT WORK (gives you a sense of what I do day-to-day) System Engineer at Mplus (DE) — building an AI-driven SaaS for residential energy efficiency analysis: B2C assessment tools plus a B2B contractor marketplace. Real users, real production. Past: R&D monitoring system for a Tel Aviv pharmaceutical client. High-traffic B2C platform with 1.2M+ users at Enkonix (AWS, message queues, cross-team integrations). ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ MY STACK AI Layer: Anthropic Claude (Tools, MCP), OpenAI Assistants & Agents SDK, LangChain, LangGraph, custom orchestration Backend: Python (primary), Node.js, TypeScript, FastAPI, Express Data: PostgreSQL, pgvector, Redis, MongoDB, Supabase Infra: AWS, Vercel, Docker, queues, webhooks, observability Web: React, Next.js (when the agent needs a UI around it) ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ HOW I WORK Day 1-2: Discovery call. I map your agent's workflow, edge cases, and failure modes BEFORE writing code. Week 1: First working agent loop in staging — you run real inputs against it. Ongoing: Async updates, weekly demo, observability dashboard so you see what the agent is doing. Handoff: Clean repo, README, evals, deployment docs, runbook for common failures. ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ NOT A FIT IF YOU NEED: ✗ "Build me a ChatGPT clone" — that's a wrapper, not an agent ✗ Pure prompt engineering with no system behind it ✗ Trading bots, gambling, or crypto-trading AI ✗ The cheapest bid — I'm not it ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ If you've hit reliability walls with n8n + GPT, or you're past the prototype stage and need a real system — send me the brief. I'll come back with a scoped architecture plan, not a generic pitch. — Martin