Skills
Product first, agents second, the stack to ship both
The first two groups are the job. The rest is what lets me do the job without waiting for someone else.
AI product management
Deciding what an agent should do, then checking it was worth building.
- Market and persona research
- Jobs-to-be-done ranked by money
- PRDs and agent briefs
- 0 to 1 roadmaps and pivots on evidence
- Eval design: offline sets, LLM judges, human review
- Agent metrics: task success, escalation rate, cost per task
- Human-in-the-loop handoff design
- Safety, guardrails and red-teaming
- Staged launches and definition of done
- Pricing and competitive teardowns
- User interviews and discovery
Agent systems
Building teams of models and running them on a budget, with guardrails.
- Multi-agent orchestration
- Agent briefs and ownership fences
- Adversarial review and persona stress tests
- Prompt and context design
- RAG pipelines
- Tool use and MCP
- Model routing by task cost and difficulty
- Token budgets and cost control
- Safe retries and rollback rails
AI and ML
- LLM integration
- Computer vision
- NLP
- Deep learning
- Generative models
- Fine-tuning
- PyTorch
- OpenCV
- Safety-critical AV systems (L3 ADAS, sensor fusion, driver monitoring)
Building and shipping
- Python
- TypeScript
- Next.js
- React
- FastAPI
- PostgreSQL
- Supabase
- Stripe
- Docker
- Vercel
- AWS
- CI/CD
- Figma
- Product analytics (Amplitude, Mixpanel)