About
Most claims on this site link to something you can check. Where one cannot be linked, the page says so.
Who
I'm Zoeb Nomi, an AI product manager at Instead, in Bengaluru. I trained as a mechanical engineer, spent more than two years and three roles building HR products at Keka, and now work on AI systems that do multi-step work in tax, and on the checks that decide when they can be trusted.
What I believe: an AI feature is not done until you can say how often it is wrong. That means three things built alongside the feature: the rules it cannot break, a record of what it did, and a count of its misses. TradeFrame and CrossSource are the two public places to see it.
Keka wrote a profile of me in Jan 2025: "Meet Zoeb Nomi: the rabbit-hole explorer and CoreHR maestro" →
-
2025–present
Instead
Product Manager · multi-step tax workflows and the evals behind them
-
2025
Multiplier
Product Manager · global employment across 150+ countries
-
2022–2025
Keka HR
Product Manager · 3 roles, 2 promotions · BGV zero-to-one to $2.7M MRR
-
2022
STOA General Management Program
-
2021–2022
Hurix Digital
Business Analyst
-
2019
MIT Aurangabad
B.E. Mechanical Engineering
Bengaluru, India
The record
Four companies, two promotions. The through-line: ship the system, then prove it works.
Sep 2025–present
Product Manager
Instead
AI-native tax research & planning platform; first new entrant to clear IRS e-filing approval alongside incumbents.
-
Workflow
Worked on a multi-step tax workflow that goes from client snapshot to strategies, memo and QC. Designed a "pipeline-collapse" eval that tests whether the chain stays coherent and honest about its own QC end to end. Cut its run time ~90%, from 10 to 15 minutes to under a minute, by re-running only the steps whose inputs changed, which turned a batch job into a tool practitioners iterate with.
-
Eval loop
Own output quality for the production tax-research LLM: flag, classify by failure mode, review weekly, on a four-dimension rubric (answerability, accuracy, citation quality, actionability) that compares model configurations and catches regressions before release. ~95% citation accuracy held on the golden eval set.
-
Pipeline
Specified the self-maintaining ingestion pipeline behind the product's answers: agentic collection, scraping, multi-model cleaning, and a daily re-scrape when a source changes. 270K+ records of primary law; a Python fetchability harness across 378 sources; ~139K documents remediated.
-
Traceability
Stopped plausible-but-wrong citations by attaching document ID and sub-type to every cited source, so any answer can be traced to the passage it stands on.
-
Onboarding
Revamped individual-filer onboarding from 30+ clicks to 8: onboarding retention 27% → 53%, paid conversion 18% → 37%.
-
Benchmarking
Scored the platform against three competing platforms across 50+ scenarios over 5 rounds; results set model and prompt decisions each round.
-
Research
Led legal-support research across 160+ tax strategies (a Legal Support Matrix from Tax Court cases, IRS rulings and audit guidance) and shipped Source Explorer, command-palette search with cross-type related sources.
Public record
What's public. The numbers above are from production. The public versions of the same work are below.
How the AI cites tax law: Research product page
instead.com
CrossSource: open harness, MIT
github.com
The judge caught a bug: essay
zoebnomi.com
Mar 2025–Sep 2025
Product Manager
Multiplier
Global employment platform enabling compliant hiring across 150+ countries.
-
Owned the Value-Added Services vertical (procurement, BGV, ITSM, partnerships) and turned it into a revenue line: $100K incremental revenue, a 47% efficiency gain from new SOPs, and the PosterElite compliance partnership launched end to end in 21 days.
Public record
What's public. The PosterElite launch was not announced publicly; the claim stands, unlinked.
Background screening for global hires: Veremark partner page
usemultiplier.com
Initiate background verification through Multiplier: help centre
help.usemultiplier.com
IT asset & equipment support at onboarding: product news
updated Aug 2025
Dec 2022–Mar 2025
Product Manager · 3 roles, 2 promotions
Keka HR
India's leading HR technology platform; $25M+ ARR CoreHR suite.
Built the Background Verification module zero to one on the Checkr API, Keka's first US-market product. It went from nothing to 28 US enterprise clients and $2.7M MRR and anchored the company's US expansion. It was a full course in integration edge cases, compliance constraints and enterprise onboarding.
-
Integrations
6 SSO integrations (Azure AD / Google Workspace) → 27% adoption lift and $121K upsell; led the Exit Module revamp → offboarding time down 28%, CSAT up 80%.
Public record
What shipped, on Keka's own surfaces. None of these name me; the profile piece does.
Background Verification module: product video
youtube.com
Keka × Checkr: marketplace listingCheckr's integration guide
keka.com · help.checkr.com
Employee Exit module revamp: announcementexit-process guide
help.keka.com
help.keka.com
Verification partners: SpringVerifyOnGridHelloVerify
keka.com
Profile: "Meet Zoeb Nomi: the rabbit-hole explorer and CoreHR maestro"
Jan 2025
Oct 2021–Oct 2022
Business Analyst
Hurix Digital
-
Led agile transformation of LMS development (43% faster delivery); drove 31% growth in customer conversions via an end-to-end video-learning rollout.
Mechanical engineering → enterprise integrations → AI workflows and the checks around them.
Public builds
Public builds
One guarded automation system and five eval harnesses, all with code.
TradeFrame: guarded automation where the AI proposes and code decides RepoGuide
CrossSource: citation accuracy in legal RAG Case studyRepoRelease v0.3Citable DOI
cheap-tier-legal-tax-hr: where a 10-cent model can sit in a regulated workflow RepoNote
screener-eval: what LLM résumé screeners actually reward Case studyRepo
mirror-eval: what AI search engines say about a person Case studyRepoRelease v1.0
decisions-api-legal-quiz: whether any confidence level makes checking safe to skip EssayRepo
Contributions
Open pull requests to other people’s eval and citation tooling. Not merged.
TaxCalcBench (Column Tax): audited all 101 test cases against the official instructions: 2 answer-key errors and 5 input conflicts found, an opt-in --accept-alternatives scoring flag, AUDIT.md in the fork PR #157 (open)
eyecite (Free Law Project): fix for Id. resolution past explanatory parentheticals (issue #348); three review findings fixed with regression tests; extraction benchmark 0 gains / 0 losses on 15,988 citations PR #360 (open)issue #361
Contact
NOW: PM AT INSTEAD
BEFORE: $2.7M MRR FROM ZERO AT KEKA HR
If you're hiring a PM for an agent or LLM product, I'd like to hear what it has to get right.
30 min · intro call