Top About Shipped Thinking Experience Contact
Product Manager · Builder — Bengaluru, India

I build products people actually use — through research, experiments, design, plain judgment, and yes, AI.

Three products built and shipped solo, plus written case studies across social, commerce, luxury, and platforms.

See the work ↗ Résumé ↗
Product Strategy ✳
Zero-to-One Builder ✳
User Research ✳
Deterministic AI ✳
Retention & Growth ✳
Consumer Social ✳
Platform Economics ✳
Execution & Delivery ✳
View Portfolio Perspective
Background

I came into product from technical delivery — SAP consulting, then running a project when the lead seat opened up. I learned to own the gap between what's promised and what ships.

Products

Three deployed products, built solo end to end, plus written studies on products I didn't build. Consumer, commerce, and a luxury service.

Principles

Judgment under uncertainty: knowing what to verify, when to abstain, and what not to build. I write down why I built what I built.

How I work across the whole job

No one project shows every part of the job. Together, these cover research, strategy, design, execution, and growth.

Shipped Products
Three products, built and shipped solo.
● In useChrome extension (MV3)
70% Week 1 Retention (n=10) · Zero Uninstalls

FitCheck ↗

ProblemIndian fashion e-commerce runs on returns, and size is the reason — the chart that would prevent the wrong buy is on the page, but nobody reads it at checkout.
DecisionBuild an ease-based verdict engine — comparing the ease a cut was designed for against the ease you'd actually get — with no LLM, because sizing is arithmetic against a published chart.
Evidence10 installs, 70% returned in week one (n=10), zero uninstalls; abstains in five defined cases rather than ship a wrong verdict.
OutcomeCaught a real sizing error that had already cost a return; a deterministic tool a user can verify against the retailer's own chart.
Role
Solo PM & builder
Timeline
Oct–Dec 2025
Platform
Chrome extension (MV3)
Stage
Shipped, in use (v0.7.0)
⚡ FitCheck Sidepanel (v0.7.0) MYNTRA · AJIO · H&M
Body Measurement: 34 in Retailer Size Chart:
Core Product Concept:
1. DOM Extractors: Content scripts inspect hydration stores on Myntra (`window.__myntra_app`), AJIO (`__PRELOADED_STATE__`), and H&M (`__NEXT_DATA__`).
2. Ease Math (No LLM): Compares body measurements against garment dimensions: 2" ease = slim, 8" ease = oversized.
3. Order History Audit: Reads `size=` params from order-history URLs to confirm if math matches past purchase choices.
PM Guardrails & Principles:
• No AI Guesses: Fits are calculated with exact math (`lib/fit-math.ts`). AI isn't used because a 1cm hallucination causes an expensive return.
• 100% Privacy: Operates entirely inside Chrome (`chrome.storage.local`). Zero network telemetry.
• Abstain Safety: Refuses to recommend if size chart data is missing or incomplete, rather than showing a wrong guess.
● DeployedConsumer AI
130+ Logged Decisions · Zero Number Hallucinations

Savio ↗

ProblemSalaried Indians have apps that track where money went, and almost nothing that answers "can I afford this, right now, without breaking something I've committed to?"
DecisionCompute every number in code and let the model write only the words; if any figure can't be verified against that computation, withhold the response rather than show it.
EvidenceCaught the model fabricating a statistic in testing and removed it from the decision path entirely; ~130 logged product decisions, including the ones reversed.
OutcomeA deployed financial companion where a wrong number is withheld, not risked — trust by construction, not by disclaimer.
Role
Solo PM & builder
Timeline
Mar–May 2026
Platform
React + Supabase web app
Stage
Deployed, actively maintained
🛡️ Savio Trust Gate (Priya Sharma) SEED: ₹98k NET INC
Monthly Surplus: ₹12,000 Impulse Threshold: ₹3,000
Core Product Concept:
1. Book-Ending Moments: Intervene during monthly rituals and windfall deposits, not aggressive checkout popups.
2. Strict Separation of Concerns: Code calculates surplus, emergency funds, and commitments; LLM only formats human explanation.
3. Named Rule Engine: Decisions cite user-configured rules by name (e.g. "Over your impulse-wait threshold of ₹3,000").
PM Guardrails & Trust:
• Trust Over Convenience: In personal finance, a wrong number destroys trust. If LLM text contradicts code math, output is suppressed.
• Buffer Protection: Automatically flags purchases that breach emergency funds or monthly budgets (Green = Safe, Yellow = Buffer Drawdown, Red = Unsafe).
● DeployedMulti-agent pipeline
13-Step Pipeline · RICE Ranked Digests

Amazon Discovery Intelligence ↗

ProblemPMs drown in review volume; the cost of skimming is invisible — you never learn what you missed.
DecisionA thirteen-step pipeline where the model runs at exactly three steps and every number a PM sees comes from deterministic code, gated by a readiness check.
EvidenceEvery digest claim cites the review behind it; unverified claims render with a visible chip; drop-accounting reports what was discarded and why.
OutcomeA month of reviews becomes a RICE-ranked, evidence-gated digest a PM can act on — with the discards shown, not hidden.
Role
Solo PM & builder
Timeline
Dec 2025–Mar 2026
Platform
TypeScript on Cloud Run
Stage
Deployed, scheduled
⚙️ 13-Step Discovery Pipeline CLOUD RUN + GEMINI
Ingest Mode: Review Audit: 140 Ingested
Core Product Concept:
1. Multi-Source Ingestion: Collects reviews from Google Play Store, App Store, and Amazon.
2. Sample/Live Toggle: A first-class switch scopes the product so curated testing data and live reviews never blend.
3. RICE & MoSCoW Scoring: Code evaluates `(Reach x Impact x Confidence x Version) / (Effort x Trend)`.
PM Guardrails & Auditing:
• No AI Subjectivity: Gemini clusters text, but math calculates priority scores to prevent hallucinated rankings.
• Discard Transparency: Drop-accounting reports exactly how many reviews were filtered out and why.
• Footnote Citations: RAG Chat answers cite exact review IDs as clickable footnotes.
Product Thinking Reddit Logo YouTube Logo
Product decisions on products I didn't build — across social, commerce, luxury, and platforms.
● Reddit Study Consumer social Growth
62% Google Organic Traffic Landing directly on Threads

Reddit Reddit: a waypoint, not a destination ↗

Problem~62% of Reddit's arrivals land on a single thread from Google, read, and leave — the activation surface is the thread page, not the homepage they never see.
DecisionReframe onboarding to live where the user already is — the thread — instead of a homepage funnel they'll never enter.
Type
Written study
Domain
Consumer social
Focus
Activation & retention
Status
Complete
📌 Waypoint Funnel Visualizer GROWTH STRATEGY
Traffic Surface: Thread Page (62%)
Strategy: Onboard inside the Google landing thread.
RESULT: +28% Activation Rate
● YouTube Study Platform economics
3-Sided Market Strategy (Viewers · Creators · Advertisers)

YouTube YouTube's Shadow R&D Lab: How Modded Apps Shape Strategy ↗

ProblemAnalytics only show what users do inside your app's rules — they miss what power users want badly enough to install third-party mods to fix.
DecisionRead community mods as a free R&D lab, then filter each signal through a three-sided market (viewers, creators, advertisers) to adopt, re-architect, or suppress.
EvidenceSponsor-skipping mods re-built into AI "Jump Ahead" heatmaps (saving creator revenue); playback controls moved to Premium; Shorts toggle suppressed to protect reach.
OutcomeA clear strategy for turning power-user workarounds into roadmap calls — without destroying the business model that pays creators.
Type
Written study
Domain
Platform economics
Focus
Shadow discovery & AI
Status
Complete
🌐 3-Sided Market Filter SHADOW R&D LAB
Evaluated Power-User Signal:
3-Step Feature Evaluation Process for PMs:
1. Separate Problem from Solution: Look past the hack to find the core desire (e.g., skip sponsor $\rightarrow$ eliminate pacing delays).
2. Stress-Test the Ecosystem: Check impact across Viewers (utility), Creators (reach & revenue), and Business (ad economics).
3. Choose Strategic Route: Adopt directly if low risk, redesign via AI if high risk but high value, or suppress if in conflict with platform moats.
Summary Insights for Interviews:
• Invisible Signal: Standard analytics miss unmet demand because product rules block the behavior from happening.
• Free R&D: Modded apps test power-user features at zero financial risk to the platform.
• Ecosystem Balance: Good PMs don't just copy user hacks; they redesign solutions to keep the marketplace sustainable.
PM Capabilities & Strategic Competencies
The frameworks, AI orchestration patterns, and growth levers I build with.
Product Strategy & Research
0-to-1 Discovery User Interviews RICE Prioritization Competitor Benchmarking
Growth & Platform Economics
Activation Funnels Retention Cohorts Multi-Sided Markets Trust Gating
AI & Technical Prototyping
LLM Integration & Orchestration Rapid Full-Stack Prototyping Deterministic AI & Grounding Browser Extension Surfaces
Work Experience
Delivery experience that carries straight into product work.
Technical Project Manager
2022 – 2025
Enterprise SAP delivery · stakeholder-facing

Delivered enterprise technical work in a high-stakes environment, then stepped into project management when a leadership gap opened mid-project. Both roles map closely to product work: translating ambiguous requirements into scoped deliverables, managing demanding stakeholders across time zones, coordinating cross-functional technical teams, and owning the distance between what was promised and what actually shipped.

Open to product roles

Let's build something worth owning.

Early-stage, India-based or remote. If you're hiring, I'd like to hear about it.

Email me ↗ LinkedIn ↗ GitHub ↗