I deploy AI systems for businesses and keep them running.
I am building an AI-assisted editor for fuel station renovation documents: client discovery, tools for each sheet and controlled revisions. I operate deployed systems and reuse solutions in Playground. Recorded demos are open without login.
- Playground: recorded demos, no login
- Mars to Table: 3D demo and calculations
- Thesis Arena: MIT-licensed code
- Thesis Arena on Devpost
- Resume and case studies
| Repo | What it shows | Key result |
|---|---|---|
| thesis-arena | Multi-agent argument stress test: progressive n8n orchestration, cited Fact Check, checkpoint retry, PostgreSQL persistence | Idea to working product in one day · OpenAI Build Week |
| youtube-car-review-pipeline | 7-workflow agentic pipeline, human-in-the-loop source selection, AI voiceover | Prep 5 h to 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 to 1 h; ~20% conversion lift, client-estimated |
| diarization-speaker-mapping-pipeline | Async diarization, deterministic Speaker-N to name mapping | 3 h to 5 min per meeting |
| voice-inventory-early-warning | Voice-driven inventory tracking + early-warning alerts | Serves 20 printers / ~60 units weekly |
Projects are published for portfolio review; client-derived workflows are sanitized and not intended for direct reuse.
Stack: JavaScript / Node.js · SQL · Python · n8n · Hermes · Claude Code · Codex · MCP · OpenAI · Anthropic · OpenRouter · PostgreSQL / pgvector · Docker · Linux · Nginx · Telegram Bot API
Contact: maxim@pmaxus.net · Resume