RailCite
Research on track.
A trust-first assistant for citing the right railway rule without inventing authority.
- 5,760 Documents indexedas of 15 Sep 2026
- 0 Invented citationsby construction
SENIOR PRODUCT MANAGER · ENTERPRISE AI · AI-NATIVE BUILDER
I turn messy problems into AI-native products people actually use.
7+ years across product, cloud, data and AI — from enterprise platforms at Godrej, Quantiphi, Shellkode and American Express to independently built AI products tested with real users.
View my work →
Ask Tushky
My AI portfolio assistant
Featured work
Real problems. Real products.
Three products that show how I turn ambiguity into something people can actually use.
How I think
A product journey, not a process.
From ambiguity to impact — six stages I return to on every product, each grounded in one real, sourced example.
01 Problem
Draft — pending sign-offFind the sentence where the current way of doing things quietly costs someone their credibility — that's the problem, not the missing feature.
See how I tested this in RailCite (opens in a new tab)“A CCI has to defend a demurrage/wharfage decision. Today that means manually walking multiple yearly PDF lists… one wrong/superseded citation damages the inspector's credibility — not the tool's.”
RailCite Discovery PRD02 Insight
Draft — pending sign-offLook for the idle time between handoffs, not the work itself — that's usually where the real delay is hiding.
See how I tested this in Nuptis → Velora (opens in a new tab)“Onboarding routinely takes 15–30 business days, yet almost none of that is active work — it is idle queue-time between cross-functional handoffs.”
Week 4 team research03 Bet
Draft — pending sign-offBet on capability over dependency: teach the reusable skill instead of doing the task for someone.
See how I tested this in TeachSpark (opens in a new tab)“Capability, not dependency.”
TeachSpark Solution-Space PRD04 Build
Draft — pending sign-offDesign the failure mode on purpose: decide what should refuse to work before deciding what should.
See how I tested this in RailCite (opens in a new tab)“Refuse is a first-class success state, never an error.”
RailCite Design.md05 Evaluate
Draft — pending sign-offRe-run the evaluation on the version of the data you'd be embarrassed to leave in, not the one that flatters the number.
See how I tested this in TeachSpark (opens in a new tab)“Activated teachers dropped from 10 to 8. Median time saved fell from 37.5 minutes to 30.”
TeachSpark build-series, Post 906 Impact
Draft — pending sign-offReport the number with its date and its caveats attached, or don't report it at all.
See how I tested this in TeachSpark (opens in a new tab)“17 teachers joined, 8 activated (47%), median 37.5 min saved (self-report).”
TeachSpark final PRD — pilot snapshot 2026-08-24, test handsets excluded
Ask
Ask Tushky
Tushar’s Portfolio Assistant
Ask anything about Tushar’s work, projects, experience, skills, product thinking, and learnings. Answers are grounded only in this portfolio.
What would you like to know?
Try asking…