I read the instruments other people ignore.
Six years on consumer products at up to 300M MAU. I find the truth hiding in retention curves and broken denominators, then ship the fix: subscriptions from zero, retention systems, and one cat nutrition app built solo, live on Google Play.
IIM ROHTAKPRE-FLIGHT · 2018 — 20 MBA · PRODUCT & STRATEGY PRE-FLIGHT
TIKTOK ADS · BYTEDANCE2020 — 21 15→60% SMB RETENTION OPEN CASE FILE →
DAILYHUNT · VERSE2022 — 26 40% TRIAL→PAID · 300M MAU OPEN CASE FILE →
Airlearn
AIRLEARN · UNACADEMY GROUP2026 — NOW +3PP D1 · A/B VERIFIED OPEN CASE FILE →
CatCulate
CATCULATE · FOUNDER2026 · LIVE LEG LIVE · GOOGLE PLAY · 80+ USERS OPEN CASE FILE →
MAU at Dailyhunt, largest platform operated
TikTok SMB, 200+ advertisers, out of the red zone
DH Premium subscription, 0→1 launch
AI content-risk detection, eval-driven
Airlearn: D1 retention +3pp — localization root-cause, metric forensics, Android push delivery
Three investigations into one number, D1 retention: rebuilding the "streak" vocabulary in 16 languages, un-poisoning the US cohort, and tracing Android push loss to the phones themselves.
Dailyhunt: DH Premium subscription 0→1 — 40% trial→paid at a 300M-MAU free platform
DH Premium from zero at a 300M-MAU free platform: found the payers hiding in the data (3× consumption → 5× LTV), priced by experiment, and built the machine that finds them.
TikTok Ads: US SMB pilot validation — advertiser retention 15→60%, evidence for 30+ country expansion
US SMB pilot validation that informed 30+ country expansion. Retention rebuilt around first-campaign success in 7 days; a platform-wide rejection bug found by walking the journey myself.
CatCulate: solo-built nutrition app, live on Google Play — deterministic engine, 600+ assertion evals
Conceived, built, and shipped an Android cat-nutrition app solo with AI-assisted development: deterministic vet-standards engine, 600+ assertion eval suite and adversarial fuzzer gating every release.
CHIKU · CATCULATE MASCOT
I ship with AI every day. CatCulate is the proof.
Claude Code is my build team: I took an idea, "is my cat actually eating right?", to a live Google Play app solo: product, engine, evals, store listing, domain. And where AI wasn't the right tool, I said no: the nutrition engine is deterministic vet-standard rules (AAFCO/FEDIAF), not an LLM, because a health-adjacent product needs explainable, testable, zero-marginal-cost answers.
Debug the denominator before the product.
A "US retention drop" was Russian VPN users polluting the cohort. The metric was sick, not the product. True US retention: 5 points higher.
Segment before you average.
The median user is a fiction. 300M users don't share one willingness to pay; Samsung and Vivo don't share one push delivery rate. Averages hide the fix.
Walk the journey yourself.
Dashboards said SMB advertisers "didn't get it." Walking signup-to-launch myself surfaced a platform bug rejecting legitimate ads. Rejections fell 60%.