I build through an AI agent. I'm responsible for every line.
"100% human orchestrating of AI — I call it 'AI in the loop' as a contrarian to the stupid 2025 'human in the loop.'"
Every repo here was built by Steve [AI], my agent. 100% machine execution. 100% human accountability.
I don't "use AI tools." I conduct an agent. The conductor takes the bow AND the blame.
→ How We Work — the full philosophy
When I consult, I find the bottleneck, build the fix, and leave. One engagement generated $200K (tax credit system, built solo in one week, error rate 30% → under 5%).
→ mordechaipotash.com — case studies, proof, engagement details
I run Steve [AI] 24/7. These are the open-source pieces:
graph TD
BM[🧠 brain-mcp] -->|memory for| AML[🔄 agent-memory-loop]
BM -->|renders to| BC[🖥️ brain-canvas]
BM -->|knowledge for| QB[📱 qinbot]
LV[🔊 local-voice-ai] -->|voice for| QB
LV -->|TTS/STT for| MN[🎙️ mordenews]
AML -->|schedules| MN
XS[🔍 x-search] -->|research for| MN
LT[🎤 live-translate] -->|uses| LV
MN -->|news for| QB
style BM fill:#e8f0fe,stroke:#58a6ff,color:#1a1a1a
style LV fill:#fff8e1,stroke:#d29922,color:#1a1a1a
style AML fill:#f0fff4,stroke:#3fb950,color:#1a1a1a
style QB fill:#f3e8ff,stroke:#8b5cf6,color:#1a1a1a
style BC fill:#fff0f0,stroke:#f85149,color:#1a1a1a
style MN fill:#9775fa,stroke:#7950f2,color:#fff
style XS fill:#ff6b6b,stroke:#e03131,color:#fff
style LT fill:#4dabf7,stroke:#1c7ed6,color:#fff
| Repo | What | ⭐ |
|---|---|---|
| brain-mcp | Cognitive prosthetic — 377K messages, 25 MCP tools, 12ms recall | 17 |
| brain-canvas | Give any LLM its own display. npx brain-canvas |
11 |
| youtube-transcription-pipeline | 32K videos → 41M words. Local ML. $0 API costs | 4 |
| agent-memory-loop | Memory maintenance layer — context windows, STATE.json, consolidation | 1 |
| local-voice-ai | Apple Silicon TTS + STT. No cloud, no API keys | 1 |
| qinbot | Full AI on a dumb phone — no browser, no apps, no images | 1 |
| mordenews | Automated daily podcast from YouTube. Zero human involvement | 🆕 |
| x-search | Search X/Twitter from terminal via Grok. No Twitter API needed | 🆕 |
| live-translate | Real-time Hebrew→English from microphone. 188 lines of bash | 🆕 |
377K messages · 25 MCP tools · 82K embeddings
9 open-source repos · 134 repos total · $200K revenue
Steve handles initial conversations. If there's fit, I engage directly.
Author: Mordechai Potash · Built through Steve [AI] · How We Work
Beit Shemesh, Israel · mordechaipotash.com · LinkedIn