AI Agentic Developer for B2B Platforms

We build production agentic-AI platforms for established European B2B companies. For the next phase of a multi-phase program — pricing automation, quote generation, LLM-based email processing, ERP/CRM integrations with human-approval gates — we're adding the developer who will carry the delivery backbone. To be precise about what we mean: NOT "AI-assisted development" — not a developer who types faster with Copilot. We're hiring an AGENTIC ENGINEER: someone who designs the loop, the skills, and the guardrails, rather than just using them. WHAT YOU'LL BUILD • Agent workflows: multi-step processes with human-approval gates (a quote never leaves without a human decision) • Skill building: structured, markdown-based skills that agents execute — turning SOPs into runnable knowledge • Loop engineering: plan → act → validate → retry → escalate; output schemas and validation layers so a hallucinated value never reaches a pricing table • MCP integrations and deterministic Python tools (FastAPI) beneath the agents: Microsoft Business Central (OData), HubSpot, Outlook/Microsoft Graph • Production LLM pipelines (Claude on Azure AI Foundry) with a data-anonymisation layer • Infrastructure: Azure VMs, Key Vault, PostgreSQL, Docker WHO WE'RE LOOKING FOR • A Claude Code power user — your daily driver, not an experiment: subagents, skills, hooks, MCP servers; you can show what you've built with it • You've shipped a production agentic loop (any stack): you know what keeps it from running away, how it validates, and when it escalates to a human • Strong Python (the language of the deterministic layer); TypeScript a plus (MCP servers) • You've shipped third-party API integrations to production (ERP, CRM, payments — any) • You work independently from written specs and acceptance tests; confident written English • Nice to have: classical ML / model evaluation / forecasting — later phases include price prediction • Also a plus: Azure/Microsoft ecosystem, Power Apps, n8n, pgvector/RAG ENGAGEMENT • Residence in the EEA is a HARD REQUIREMENT (client data-processing constraint) — please don't apply from outside the EEA, we cannot make exceptions • Long-term, multi-phase program, ~40 h/week from September (part-time warm-up possible earlier); honestly stated: strong pipeline, no guaranteed annual volume • Selection includes a short PAID trial task (4–6 h at your offered rate) — your time is money too • You work directly with a senior AI architect (10+ years in AI, PhD): ready PRDs, per-phase specs, acceptance tests, and a working agentic dev environment; everything runs in the client's own tenant, GDPR-aware by design • This is a live, daily-used production system — not a demo HOW TO APPLY — read this or your proposal will be skipped: Start your proposal with a short paragraph on the best agentic system you've built: what the loop did, where the validation sat, where the human gate was, and what stopped it from running away. GitHub/portfolio links welcome. Generic proposals and agency templates are declined without reply.