Now
What I'm shipping and thinking about, updated regularly. /now page convention.
Shipping
- ResumeForge — AI resume-optimisation POC. Turns a master resume + job description into an ATS-optimised two-page Word doc, gated by a five-question discovery loop. Repository.
- ai4pdf v1.0 — full-stack AI PDF tool: edit, scan, OCR, merge, split, sign, annotate, chat/summarize. 100% client-side. Repo.
- BeBuddy AI v1.0 — private wellness PWA (stress · sleep · mood · routines). On-device, no medical advice. Repo.
- pm-os v1.0 — PM operating system, PRD → SDLC via Claude Skills + MetaGPT + Promptfoo. Live app.
- agentic-compliance-evals (in flight) — open eval harness for HITL AI in KYC/AML + clinical CDS. Promptfoo-compatible, 50 golden cases.
- Portfolio README v2 — every case study migrating to Problem → Approach → Trade-offs → Outcome → Failure structure.
Thinking about
- Why calibration matters more than correctness in regulated AI evals.
- How HITL gates should differ between BFSI (low latency, high volume) and clinical CDS (low volume, high stakes).
- Cost-quality-latency trade-offs when routing between Claude, GPT-4o, Gemini for compliance workloads.
Reading
- NIST AI RMF and its enforcement patterns in BFSI.
- FDA guidance on AI/ML clinical decision support.
- Anthropic model card + safety evals as a reference schema for regulated deployments.
Open to
- AI PM · Senior PM · AI Platform PM roles in regulated AI (BFSI, healthcare, insurance).
- Reviewing PRDs / eval designs for HITL agentic products.
- Collaboration on open evals — see GitHub.
Last updated: 2026-07 · Home · Writing