I design, build, and run production AI automation systems end to end: from the first client interview to a deployed, self-hosted pipeline kept in operation.
Focus: multi-step agentic workflows on n8n — human-in-the-loop control, strict input/output contracts, managed token cost. Own infrastructure: DigitalOcean · Docker · n8n · PostgreSQL · NocoDB — ~30 pipelines in operation.
| Repo | What it shows | Key result |
|---|---|---|
| thesis-arena | Multi-agent argument stress test: progressive n8n orchestration, cited Fact Check, checkpoint retry, PostgreSQL persistence | Live GPT-5.6 product · OpenAI Build Week |
| youtube-car-review-pipeline | 7-workflow agentic pipeline, human-in-the-loop source selection, AI voiceover | 5 h → 5 min per review; 130 cars |
| dcp-consultation-assistant | Telegram AI agent, per-request Postgres memory, explicit human escalation | −60% first-line requests |
| design-brief-visualizer | Human-in-the-loop style confirmation before expensive image generation | Brief: 2 days → 1 h; +20% conversion |
| diarization-speaker-mapping-pipeline | Async diarization, deterministic Speaker-N → name mapping | 3 h → 5 min per meeting |
| voice-inventory-early-warning | Voice-driven inventory tracking + early-warning alerts | Serves 20 printers / ~60 SKUs weekly |
Projects are published for portfolio review; client-derived workflows are sanitized and not intended for direct reuse.
Stack: n8n · Python · JavaScript · SQL · PostgreSQL/pgvector · Docker · Linux · Nginx · OpenAI · Anthropic · OpenRouter · Telegram Bot API · MCP
Contact: maxim@pmaxus.net · Telegram @inorout · Personal site with resume https://pmaxus.net/resume/ai-automation-engineer Case studies and references on request