A real-time coaching layer for B2B sales calls. The rep gets the next best move on screen; the prospect never sees the software.
- Live context — CRM, email, past calls, and the VP Sales playbook become useful during the conversation, not after it.
- Invisible overlay — protected from screen sharing across Zoom, Google Meet, Microsoft Teams, Loom, and OBS.
- Latency as product — Recall.ai → Gladia ASR → Anthropic streaming, engineered toward a sub-2s p95 response.
- A learning loop — manager corrections become governed rules that improve the next call for every rep.
Shot is the system I want to build for the next decade. The source is private; the public project brief documents the product and architecture.
I use Claude, Cursor, Codex, Supastarter, and specialized agents as an engineering team. The interesting part is not pretending the code was typed alone. It is designing the system, giving agents the right context, setting the constraints, and proving the result.
signal → context → plan → agents → gates → evidence → shipped system
That operating model is becoming DevFactory Core: a proof-first orchestration layer with reusable skills, durable loops, bounded workers, model routing, and evidence gates. A clean public core is in preparation.
| System | What it proves |
|---|---|
| email-pipeline | A focused, cache-aware enrichment pipeline built to turn expensive external calls into a repeatable data system |
| Closer Claw | A full competitive-intelligence SaaS: adaptive scraping, background workers, AI analysis, billing, and multi-tenant product infrastructure |
| Shot | Product thesis and architecture for real-time AI coaching; the production source remains private |
| Project | System | Status |
|---|---|---|
| Shot | Real-time AI coaching for B2B sales teams | Building now · source private |
| email-pipeline | Autonomous, cache-aware B2B email enrichment | Live |
| Closer Claw | Competitive intelligence with selectors that survive DOM changes | Archived after the pivot |
| DevFactory Core | Proof-first development factory for AI coding agents | Sanitizing for public release |
signal → context → decision → action → measured feedback → better rule
I work end-to-end: product, distributed systems, AI pipelines, desktop software, data, and deployment. My bias is toward short feedback loops, explicit failure modes, and evidence before claims.
