A decade of operator work, now teaching the patterns.
I've shipped product across six industries: AI-native consumer in fashion and content creation, last-mile B2C logistics in an emerging market, regulated SaaS in pharma, vertical ERP for an underserved industry, and outcomes-driven ed-tech.
The through-line: whether real users adopt what we ship. Not just launches. Not just metrics. The unglamorous part — does the product earn its place in someone's actual day?
I work in AI now. I think about agents, evals, harness architecture, and the design choices that make AI products useful instead of just impressive on a demo reel. What I publish here is a framework I actually use, a pattern I've seen work, or a demo I've actually built.
Specifically, four kinds of work.
Cutting AI roadmaps down to the version that actually ships — and earning the right to build the next one.
Building eval suites for tasks where there is no obvious ground truth and the deadline is real.
Going deep on one industry’s actual operating model — not the version the customer first describes.
Specs that survive contact with reality. The PM ↔ design ↔ engineering loop, in both directions.
The frameworks worth handing forward.
- →Eval design when ground truth is fuzzy
- →Scoping AI features without overengineering
- →First-10-customers playbooks for B2B AI
- →Vertical SaaS distribution motions
Email is fastest: ruchitdalwadi001@gmail.com. Specific notes about what you're building get faster replies than generic ones.