Vaishnavi Upadhyay

AI Platform Product Manager

6+ yrs

Product, platform & governance

500+

Enterprise partners served

160+

Developer teams enabled

Scroll

VBot

Ask VBot anything, in my voice

A small grounded agent that replies in casual paragraphs, the way I would - and honestly says when it doesn't know.

Hey, I'm VBot. I answer the way Vaishnavi would - casual, in plain paragraphs, grounded in what she has actually shipped. Ask me about agent design, platform work, or how she prioritises.

Try asking

Coverage Tracker

Where I hear about it, mapped

Google AI Garage sits in the middle; every source I follow orbits it. Hover a point to read the detail, click to open the source.

Google AI Garage

linkedin.com hcamag.com linkedin.com hrtoday.in linkedin.com

linkedin.com

VP People Products on AI in people operations

A leadership post on how people products are being rebuilt around AI - the clearest signal of where internal agent work is heading.

Rahul Jindal

linkedin.comVP People Products on AI in people operations linkedin.comBuild updates straight from the feed linkedin.comHiring threads show what the team is really building

Articles

hcamag.comGoogle's AI-driven transformation of people operations hrtoday.inLearning & development leadership moves in India

The Playbook

My Agentic Workforce Playbook

A field guide for building AI agents for a large, distributed internal workforce.

Medium article

The AI Agent Playbook

01Start with the decision
02Agent taxonomy
03Designing for trust
04Evaluation
05Rollout
06Governance
07Metrics

Read the full playbook

Fun

Skill Catch, played with your hands

Balloons drift down carrying the product-manager qualities you look for. Hold an open hand in front of your camera to move the ring at the centre of your palm, then close your fist to grab whichever ones matter to you. Stop any time - your picks are sent to me as feedback.

How to play

1Allow camera access. Everything runs on your device; no video leaves the browser.
2Hold one open palm up, roughly an arm's length from the screen. Your fingers are drawn linked, with a ring at the centre of your palm.
3Close your fingers into a fist over a balloon to catch it. The ring turns green while you are gripping.
4Balloons start six seconds after you hit start. Catch 10 qualities, or press stop whenever you are done. No camera? Move your mouse and hold the button instead.

0 / 10 caughtOpen hand

Add your name and email, enable the camera, then start catching.

Fun

Motion as an interface

A small experiment in camera-based interaction that the Skill Catch game below is built on.

MotionJump uses the webcam to turn body movement into game input - no controller, no keyboard. Pose landmarks are tracked frame by frame, smoothed, and mapped to a jump when the player physically leaves the ground.

The same idea powers Skill Catch: an open hand moves the ring, a closed fist grabs. It is a nice reminder that the fastest way to explain an interaction model is to let someone play with it.

View MotionJump on GitHub

Ideas

Six agents I would build next

Each one is a full write-up: the problem, the agent design, the guardrails, and the metrics I would hold myself to.

Evaluation & Quality

Multilingual Evaluation Mesh

One evaluation fabric that proves an assistant is as good in Tamil, Swahili and Portuguese as it is in English.

0 to 1, two quarters

Governance & Policy

Cross-Border Compliance Agent

Reads a feature spec and tells you, per market, what you can ship, what you must change, and what needs counsel.

0 to 1, one quarter

Language & Access

Low-Resource Language Bootstrapper

Gets a product to usable quality in a language with almost no digital corpus, in weeks not years.

0 to 1, two quarters

Vertical Studios

Vertical Studio Agent Kit

A shared substrate that lets a five-person vertical team ship a production agent in a fortnight.

Platform, ongoing

Trust & Safety

Agent Incident Responder

When an agent misbehaves at 3am in a market you do not speak, this is what finds it and contains it.

0 to 1, one quarter

Partner Relationship Management

Partner Onboarding Agent

Turns a six-week, multi-jurisdiction partner onboarding into a supervised three-day loop.

0 to 1, one quarter

If you hire me

My first 90 days and how I build agents

A concrete plan, plus the execution process I use to take an agent from a vague ambition to something governed, measured and running in production.

Days 1 to 30

Listen, instrument, and find the truth

Understand the system before touching it

Sit in on twenty real user or partner sessions and write the friction log myself, unfiltered.
Read the last two quarters of incidents, evals and escalations; they describe the product more honestly than any roadmap.
Map every agent surface already in production and whether anyone can measure it.
Publish a one-page problem inventory ranked by cost, not by who complained loudest.

Outcome: A shared, evidence-backed view of what is actually broken.

Days 31 to 60

Prove one loop end to end

Ship a narrow agent with real guardrails

Pick the highest-cost, most rule-dense workflow and scope it to one supervised loop.
Stand up the eval harness before the agent: golden set, graders, and a release bar.
Ship behind a human-approval surface so every write action is reviewed as a diff.
Instrument reviewer edit rate, grounding rate and time saved from day one.

Outcome: One agent in production with a defensible quality story.

Days 61 to 90

Turn the loop into a platform

Make the second agent cheaper than the first

Extract the reusable pieces: tool registry, approval surface, trace store, eval runner.
Write the agent design review checklist so other teams reuse the guardrails, not just the code.
Hand the second use case to another team and coach rather than build it.
Set the operating cadence: weekly eval deltas, monthly risk review, quarterly capability bets.

Outcome: A repeatable path from idea to governed agent.

Frame

Write the problem as a job, a cost and a decision. If I cannot state what decision the agent is making on someone's behalf, it is automation wearing agent clothing.

Constrain

Decide the blast radius first. What can the agent read, what can it write, and what always needs a human. Constraints defined late become incidents.

Measure

Build the eval harness before the agent. A golden set, rubric graders calibrated against humans, and a release bar that can actually block a launch.

Ship narrow

One workflow, one persona, one surface, with human approval in the loop. Narrow scope is what makes the quality signal readable.

Watch

Reviewer edit rate, grounding rate, escalation rate, and cost per resolved task. Four numbers a leadership team can hold in their head.

Widen

Only remove human approval where the data earns it, one claim type at a time, with a rollback that takes minutes.

Community

Google Product Expert

Platinum contributor in the Google Product Expert Program, supporting real users since 2019.

Platinum · Since 2019

Seven years inside the user side of Google products

Part of the Google Product Expert Program, where I help users across markets with Assistant, Search and Translate - and feed the recurring failure patterns back to product teams. It is the closest thing I have to a permanent user research loop: it is where I learn how features actually break for people in other languages, other devices and other network conditions, long before that shows up in a dashboard.

Google AssistantGoogle SearchGoogle Translate

Case Studies

Things I actually shipped

Problem, approach, what I did, outcome - and the part most decks leave out: what I'd do differently.

About

Product thinking, end to end

Product Manager with 6+ years of experience driving quality, risk, and governance outcomes for AI-driven, partner-facing platforms at enterprise scale. I build measurement frameworks, telemetry-based risk visibility, and automated governance checks - turning ambiguous, cross-functional problems into scalable, data-driven products.

Adventurous outside of work

Trails, long drives and new cities on weekends - the same curiosity I bring to product work.

Years in Product

500+

Enterprise Partners

160+

Developer Teams Enabled

40%

Manual Config Reduced

Resume

Six years of building

Browse the full track record below, or take a copy with you.

Download Resume ↓

Product Manager

Aug 2022 - Present

Zoho Corporation Pvt Ltd · Chennai, India

Own governance, compliance, and role-based access control frameworks for Partner Relationship Management integrations across 500+ enterprise partners, improving partner onboarding efficiency by 30%.
Lead AI strategy and feature delivery for Zia, Zoho's AI assistant, using telemetry, usage analytics, and A/B experimentation - reducing manual configuration by 40%.
Own the roadmap for an AI platform and self-service developer studio empowering 160+ developer teams to launch vertical apps and CRM extensions through APIs and SDKs.
Partner daily with distributed Engineering, UX, and security teams to translate ambiguous platform and data-risk problems into reliable, scalable operational tools.
Brief senior leadership on platform strategy, risk posture, and migration roadmaps, securing executive buy-in for solutions built at global scale.

Member of Technical Staff - Product Quality

Aug 2018 - Aug 2022

Zoho Corporation Pvt Ltd · Chennai, India

Directed quality management and risk validation strategy for the Zia AI platform's integration pipelines, increasing deployment success rates by 20%.
Managed technical lifecycle and integration architecture across 5+ Zoho products for secure, high-throughput platform features.
Partnered with Engineering and UX on user research to map Critical User Journeys and surface adoption risks in high-stakes data integration flows.

Software Engineer in Test (QA)

Jul 2017 - Jul 2018

Caratlane Trading Pvt Ltd · Chennai, India

Performed business analysis and structured risk/quality testing for the Titan Eyeplus e-commerce platform, defining test methodology and scope.
Drove new product development with a 15-person developer team on releases contributing to a 25% month-over-month revenue increase in FY19.

Prefer the PDF? vaishu09up@gmail.com · +91-8124428719

Vaishnavi Upadhyay

Ask VBot anything, in my voice

Where I hear about it, mapped

VP People Products on AI in people operations

My Agentic Workforce Playbook

The AI Agent Playbook

Skill Catch, played with your hands

Motion as an interface

Six agents I would build next

Multilingual Evaluation Mesh

Cross-Border Compliance Agent

Low-Resource Language Bootstrapper

Vertical Studio Agent Kit

Agent Incident Responder

Partner Onboarding Agent

My first 90 days and how I build agents

Listen, instrument, and find the truth

Prove one loop end to end

Turn the loop into a platform

Google Product Expert

Seven years inside the user side of Google products

Writing I publish in public

Summers and Product Management

Mahakumbh 2025 · Brand strategy at scale

JioCinema × Disney+ Hotstar

FeatureLens #3 · Spotify

FeatureLens #2 · WhatsApp

Ola vs Uber

Google Lens & video search

Things I actually shipped

Product thinking, end to end

Six years of building

Product Manager

Member of Technical Staff - Product Quality

Software Engineer in Test (QA)