rd.
Ruchit Dalwadi

I help operators ship AI-native product.

Frameworks, evals, and shipping playbooks from a decade of operator work across six industries — AI-native consumer, regulated SaaS, emerging-market logistics, vertical workflow software, and outcomes-driven learning.

/ Track record

Six industries. One through-line.

6
industries shipped across
10+
years operator work
millions
of end-users reached
/01 · AI × Content editing

Reducing post-capture friction for creators

AI-in-the-editor only works when the AI's suggestions are *commitable in one click* and *reversible in one click*. Adding confidence scores didn't matter. Adding "accept all / accept this / undo" did.

/02 · AI × Fashion

Personalized discovery for a visual-first category

In taste-driven categories, the recommendation engine has to be *visibly considerate* — not just accurate. Users forgive a wrong suggestion if the system shows it understood why they might like it. Black-box "for you" feeds train distrust.

/03 · Ed-tech

Outcomes-driven learning at scale

Students don't drop out because the content is bad. They drop out because the next step isn't obvious. Removing decisions from the journey — exactly one thing to do today — moved completion more than any pedagogy change.

/04 · B2C last-mile logistics

Operating-tech for dense urban last-mile delivery

In emerging-market logistics, the product *is* the cost structure. Every feature has to survive a $5/day operator running it on a $50 phone with 200MB data. Anything that doesn't pay for itself in offline reliability gets cut.

/05 · Pharma content SaaS

Editorial workflow under regulatory compliance

In regulated SaaS, the audit trail is the actual product. The editor is just the UI on top. Once we built around "every change is provable, attributable, and reversible," everything else became simpler — including the editor.

/06 · ERP SaaS

Workflow software replacing spreadsheets for a specific operator

For vertical SaaS, the onboarding *is* the product. A customer who can't get their existing data into the new system doesn't churn — they never start. We built the migration as a concierge service before we built half the features.

/ Writing

Frameworks, not hot takes.

How to evaluate LLM outputs without ground truth

AI · 4 min read

“Stop hunting for a labeled dataset. Build evals out of signals you already have, and treat the eval suite as a product that ships alongside the model.”

Jun 2026

First ten customers for a B2B AI startup

Startups · 4 min read

“The first ten customers of a B2B AI startup aren’t bought with marketing — they’re recruited with hand-built integrations in exchange for design-partnership rights.”

Jun 2026

What I learned shipping product across six industries

Product · 4 min read

“The skill that transfers across industries is modelling the user’s actual job at a level of specificity no one in the room currently has. Everything else is local.”

Jun 2026
/ The through-line
“Get to the specific job. Then build the most boring version of the product that does it well.”
from ‘What I learned shipping across six industries’