Mayank Singh Raghav — AI Product Manager

SHIPPING: SUNSET — WEEK 3 OF 8

PRODUCT / AI / 0→1

[ SCROLL TO SEE THE EVIDENCE ↓ ]

MAYANK

SINGH

RAGHAV

PRODUCT THINKER.

AI BUILDER.

PROBLEM OBSESSED.

EXPLORE THE LAB ↓DOWNLOAD RÉSUMÉ (PDF) →

"I turn ambiguous problems into products, experiments and measurable outcomes."

EXPLORE THE LAB ↓DOWNLOAD RÉSUMÉ (PDF) →

// PROFILE

POSITION: AI Product Manager · PM-engineer hybrid
BASED IN: Gurugram, India
EDUCATION: Amity · NextLeap
STATUS: AVAILABLE FOR 0→1

⚡ NOW IN THE LABNOW BUILDING: SUNSET — MULTI-AGENT FEATURE DEPRECATION · WEEK 3 OF 8✦

[01]

WHO — IDENTITY & STACK

PM-ENGINEER HYBRID

"I turn ambiguous problems into products, experiments and measurable outcomes."

I operate at the intersection of technical architecture and product strategy. Having led edge computer vision products and authored comprehensive AI copilot PRDs, I bring an engineer's technical depth alongside a product manager's discipline for user discovery, prioritization, and unit economics.

// CREDENTIALS

▪B.Tech Computer Science — Amity University (2019–2023)

▪NextLeap PM Fellowship

// STACK

Production tools, frameworks, and architectures used across lab builds:

PythonFastAPIDjangoLangChainRAG pipelinesChromaDBComputer VisionEdge computing

THE ROADPRODUCT & ENGINEERING ROLES

Aug 2025 — Apr 2026

WCO Global / Elle Global

Product & Technical Lead — Football Analytics

Owned end-to-end product lifecycle for a multi-camera football analytics system shipped on edge hardware.

→Authored PRDs and defined specs for RTSP streaming and audio/video sync.
→Coordinated hardware vendors and engineering teams to ship on Raspberry Pi 5 edge devices.
→Diagnosed latency bottlenecks via data analysis and drove engineering fixes for maximum reliability.

May — Jun 2025

Bluestock Fintech

Software Developer Intern (PM-adjacent)

Mapped customer journey for IPO data flows on a digital fintech platform.

→Prioritized API requirements with stakeholders, delivering 40% faster response times.
→Built analytics dashboards (PostgreSQL, Matplotlib) for end-user financial insights.

Jan — Feb 2025

Excelerate

Associate Project Manager Intern

Ran a 6-member cross-functional team on Agile/Scrum with full backlog ownership.

→Led daily standups, JIRA backlog prioritization, and sprint planning.
→Achieved 100% on-time delivery across the client engagement.

Sep — Dec 2024

CodeCore Global

AI Product Intern

Scoped and shipped an AI chatbot for rural communities from discovery to QA.

→Ran customer discovery around agricultural schemes and government subsidies.
→Owned design-to-QA delivery using perplexity-based success metrics.

[02]

THINK — GIVE ME A PROBLEM

Type any product problem — or pick one below. It gets broken down, live, into USER → PROBLEM → HYPOTHESIS → MVP → METRIC.

TRY:

BREAKDOWNSpotify × AI Discovery

1
// USER
Avid music listeners looking for niche genre exploration.
2
// PROBLEM
Algorithmic playlists repeat comfort tracks; listeners struggle to discover truly novel subgenres.
3
// HYPOTHESIS
Natural language contextual prompts + audio spectrum analysis unlock higher discovery satisfaction.
4
// MVP
Streamlit UI wrapping Spotify Web API + Gemini for conversational radio station generation.
5
// METRIC
40%+ save rate for recommended tracks.

// THE METHOD — the six steps behind every breakdown

01USER PROBLEM
02DATA?
03RESEARCH / TEST
04SHIP
05MEASURE
06ITERATE

A ranked list of twenty tools isn't a recommendation. It's the same decision, handed back to you with more homework.

— on building TROVE

[03]

BUILD — THE LAB (EXPERIMENTS 001–005)

METRICSSHIPPED & SIMULATED OUTCOMES

[01]

6,800

REVIEWS PROCESSED

By one LLM pipeline (Groww PulseAI)

[02]

PRODUCTS IN THE LAB

Active experiments & specs

[03]

PRODUCT HUNT LAUNCH

Trove community launch

[04]SIMULATED

+78

NPS LIFT

Simulated outcome (Groww PulseAI)

FLAGSHIP

TROVE

LIVE · LAUNCHED ON PRODUCT HUNT

// THE PAIN

"Best tools" lists give you 20 ranked links and no decision.

// THE DECISION

Does one confident, hand-checked recommendation (with reasoning + 2 alternatives) beat a ranked list?

↳ THE FORK: I could have auto-scraped a ranked list of twenty tools; I chose one hand-checked pick with its reasoning because a ranked list defers the decision and a recommendation makes it.

// WHAT I BUILT

Curated tool-recommendation product; every entry hand-checked with a stated reason for inclusion.

// WHAT HAPPENED

Launched on Product Hunt and Reddit; users specifically praised the hand-checked/curated angle over auto-scraped directories.

OPEN TROVE ↗

EXPERIMENT 001

GROWW PULSE AI

STATUS: LIVE

// PROBLEM

~6,800 app store reviews. Nobody had time to read them.

// HYPOTHESIS

Can an LLM pipeline turn thousands of raw reviews into ranked, actionable product insights?

// WHAT IT IS

Deployed LLM-based review intelligence pipeline.

// RESULT / STATUSSIMULATED

+78 SIMULATED NPS LIFT

↳ THE FORK: I could have hand-sampled a few hundred reviews; I chose to run the full ~6,800 through an LLM pipeline because the signal I needed lived in the volume, not a sample of it.

OPEN EXPERIMENT →

EXPERIMENT 003

SUNSET

STATUS: IN PROGRESS

// PROBLEM

Mature B2B SaaS teams can't kill features safely — usage, contract, and support risk live in different silos.

// HYPOTHESIS

Can a multi-agent system (usage auditor + contract-risk auditor + support-signal auditor) reconcile into a single 'kill memo' with at-risk accounts and a migration plan?

// WHAT IT IS

Multi-agent feature-deprecation tool for Product Ops / Platform PMs. Entire stack is open-source/free-tier by design.

// RESULT / STATUS

Solo build in progress, 6–8 week timeline.

↳ THE FORK: I could have pointed one model at the whole problem; I chose three specialized auditors — usage, contract, support — reconciling into a single kill memo, because those are different risks that shouldn't share a prompt.

INTERNAL SPEC / IN BUILD

EXPERIMENT 004

BLAST RADIUS

STATUS: SPEC

// PROBLEM

Pricing and packaging changes ship blind — nobody knows which contracts and renewals they detonate.

// HYPOTHESIS

Can a contract auditor + renewal-risk auditor + deterministic simulation engine predict the blast radius of a pricing change before it ships?

// WHAT IT IS

Multi-agent pricing/packaging change simulator with a human-in-the-loop approval gate for migration comms. Full build spec written.

// RESULT / STATUS

Full build spec written.

↳ THE FORK: I could have fully automated the pricing simulation; I chose a deterministic engine with a human approval gate on migration comms because a wrong blast-radius call ships straight to real customers.

INTERNAL SPEC / IN BUILD

EXPERIMENT 005

HEATWAVE VULNERABILITY MAPPER

STATUS: IN PROGRESS

// PROBLEM

Heat kills unevenly, but city responses treat wards identically.

// HYPOTHESIS

Can Landsat thermal data + Census demographics produce ward-level vulnerability scores a city could actually act on?

// WHAT IT IS

Geospatial data product scoring ward-level heatwave vulnerability.

// RESULT / STATUS

Geospatial pipeline & scoring model in progress.

↳ THE FORK: I could have scored heat risk from temperature alone; I chose to fuse Landsat thermal data with Census demographics because heat kills unevenly, and a score a city can act on has to show who's exposed, not just where it's hot.

INTERNAL SPEC / IN BUILD

[04]