Shreya Ghosh
Approach Experience Projects About
AI-native product manager · Bengaluru

Most product problems are behavioural before they're technical.

My psych degree helps me understand people. AI is part of how I research, make decisions, test ideas, and ship faster.

Built insideGravity AI·Vibe AI·DeepThought
Shreya Ghosh
“The person I would trust to red team my thoughts and give really good insights on spot.”
— Aryan Agarwal, Founder, Gravity AI
My best work 01 / 04

Some high impact work.

01 Vibe AI
Turns out people didn’t want an AI companion at all. So Vibe pivoted.

My research surfaced it, and it defined a major pivot for the product.

02 Gravity AI
I picked the north star metric the whole company chased.

Then every team got their own metric feeding into it — and I built the acquisition plan around it.

03 Gravity AI
Set a custom sprint Gravity’s engineers still ship on today.

No cadence existed, so I built one — based on agile methodologies. It stuck

04 Gravity AI
600 builders get the right message without anyone sending it.

I replaced manual SMTP campaigns with triggers that fire off what people actually do.

Inside an AI-native approach

From messy problems to testable systems.

The team asked me
standup
“Make it sound human na Shreya.”
We were building the most human-like AI friend.
How I executed
Evals first
Defined what “human” meant and turned subjective preferences into observable criteria.
GPT and Claude for model research
Read model docs, cherry picked 3–4 models.
OpenRouter for A/B tests
100+ custom prompts across DeepSeek, Gemini, Mistral, and even open-source — benchmarked for output.
“Sounds human” became something we could measure, compare, and improve.
The team asked me
standup
“How can we make it easier for AI builders to understand the platform?”
Builders were dropping off and nobody knew where.
How I executed
PostHog MCP + Claude
Plus a little custom SQL to understand builder churn points.
Customer.io campaign
Scheduled interviews and ran UX interviews.
Synthesised insights into sheets
Identified where the product and users’ mental models diverged. Mapped the failure points.
Translated the findings into a journey spec
Gave design an evidence-backed flow to execute.
An unclear drop-off became a specific product problem the team could solve.
Some strong opinions
Experience

Two AI startups. At both, I set up a function that didn't exist yet.

Gravity AI

Growth Product Manager
Feb – Aug 2026 · Bengaluru
AI agent marketplace
Promoted from Founder's Associate

I joined with no growth function in place, so I set one up — along with the research function. I defined what activation meant, then built the systems that moved it.

  • Ran churn research across 17 journey touchpoints with lapsed builders, found the primary drop-off, and drove the fix into the UI roadmap.
  • Replaced manual SMTP campaigns with a behavioural lifecycle system in Customer.io, triggered from Supabase — including re-engagement for builders who claimed a task and went quiet.
  • Defined the north star metric, gave every team a metric that laddered into it, and ran the acquisition strategy it shaped with marketing.
  • Shipped a reusable competitor-research agent with a validation and dedup layer, which removed the manual review step entirely.

Vibe AI

Founder's Associate, Product & Research
Aug 2025 – Feb 2026 · Bengaluru
Consumer AI companion
Promoted from Research Intern

I owned the loop between what users felt and what the model did — evaluation, experiments, and the community that fed both.

  • Designed the model evaluation: benchmarked candidates against a fixed prompt, A/B tested 100+ variants, and turned "the conversation feels off" into specific fixes that shipped.
  • Recruited and ran a 200-person beta. Findings shipped into onboarding, conversational realism and engagement.
  • Read up on anthropomorphism and turned it into a position the team could argue with. It changed the companion's conversational direction.
  • Taught myself context engineering from zero to do the eval work properly, rather than handing it to an engineer.

DeepThought EduTech

Micro Lead
2024 – 2025
Promoted from Org Development Intern
  • Led a five-person team running training and performance workflows. I was 20.
  • Owned stakeholder outreach — built the email campaign and cold-called 40 prospects myself.
  • Coached a colleague on communication. A client hired him after the training.
Selected work

When I feel stuck at work, I build.

It answers the question I can’t answer from inside one company: what would I do here if this were a different product?

Thinklet preview
Thinklet · docked side panel, live on the Chrome Web Store 01 / 02
Possibly preview
Possibly · from no idea to your first ten customers 02 / 02
What didn't ship
Mind Mess

An early-stage mental wellness content venture I founded solo, 2024–2025. 3K+ views in the first week — my first go at building an audience from nothing, and where I learned how much a voice matters before a product does.

Read the article
PM Bot for Slack

An AI bot that lived in my Slack and helped me sharpen my product sense. It taught me to scope an MVP properly instead of building every request at once.

Read the post
About

I keep finding myself in unknown places. Caught a few on cam.

I’m a Bangalorean who was born in Delhi and ran to Kolkata for college. I know. Absurd.

But so has been my work life — tutoring since I was 17, trying to build a startup, leading a team at 20, abandoning my plans for a master’s in psychology, and teaching myself how to ship products with AI.

Phew. It’s been one ambiguous ride.

But that ambiguity has also led me to some of the most interesting work I’ve done — across product and growth, only with the slightly weird lens of someone who studied psychology first and figured out the rest along the way.

Anyway, two truths and a lie: I speak Japanese, I love coffee, and I’m a Plat in Valorant. I’ll tell you the answer when we connect.

AI Agent Builder Series 2026 opening slide, powered by Google and AI House
2026 · Experiments with Google ADK.
Conference badge reading Shreya Ghosh, Founder, Mind Mess
2024 · First real networking event. Had to Google “what does scalable mean?”
Shreya with her cohort in a Bengaluru office, one person holding a sign reading If it moves the metric, do it now
2025 · Experimenting with growth.
GrowthX AI Festival signage by a pool in Goa
2026 · Learnt from 300 awesome people about growth, product and more.
Claude Code
Claude Code
Figma
Figma
Codex
Codex
Framer
Framer
Supabase
Supabase
Ollama
Ollama
PostHog
PostHog
Linear
Linear
SQL
SQL
Jira
Jira
Firecrawl
Firecrawl
Activepieces
Activepieces
Customer.io
Customer.io
Apollo.io
Apollo.io
Hermes Agent
Hermes Agent
Higgsfield
Higgsfield

My tools

Contact

Great products start with good conversations.

Let’s have one.

ghoshshreya852@gmail.com
LinkedIn X
Shreya Ghosh
2026 · AI and Growth Product Manager