There's a new gap in product organizations: teams have LLM ambitions, engineers experimenting with models, and executives asking "what's our AI strategy", but nobody owns the product judgment in between. Which use cases justify probabilistic software? What does "good enough to ship" mean when outputs vary? Who decides what happens when the model is wrong? That ownership is what I provide.
A PM who ships AI
I co-founded WisOwl AI and built its core systems personally: an embedding-based semantic matching engine on FAISS and Supabase pgvector, and autonomous recruiter agents matching supply and demand in the Indian hiring market in real time. The platform reached 10,000+ organic jobseeker signups and 10+ recruitment-firm and startup customers with zero paid marketing. Before that, eight years at CaaStle running consumer experience and growth product across a ~$50M+ ARR portfolio. So the AI work I do always answers to retention and revenue.
AI product management services
- AI product strategy: a candid map of where LLMs earn their place in your product and where they only add cost and risk, with build-versus-API decisions, model selection logic, and unit economics attached.
- RAG and search products: scoping and product-managing retrieval systems: corpus strategy, retrieval evaluation, and the UX for the queries retrieval can't save.
- Agentic products: deciding when autonomy is justified, designing guardrails and human-in-the-loop checkpoints, and defining evals before the build rather than after the incident.
- AI feature delivery: embedded fractional product leadership, 1 to 3 days a week, running your AI roadmap from PRD through launch, including the evaluation harnesses that let you measure quality.
- AI product audit: a one-to-two-week review of an existing AI build: retrieval quality, eval coverage, failure UX, latency and token economics, ending in a ranked fix list.
The pages linked below go deeper on each. If you already know which problem you have, book the call and skip the reading.