Skip to content
Departures boardOpen to roles · Remote or onsite

AI Product Manager · 8 years · AI agents, GenAI, eCommerce SaaS

Eight years across AI product, eCommerce SaaS and large-scale delivery, including 3+ years owning AI and GenAI products and leading teams of up to 23 across engineering, AI and QA. I prototype against real models before writing a spec, then work with engineering from something that already runs.

Get in touchDownload CV
LatestAkshar Bench, a blind human eval of Indian-script text in AI image models
75 days
REALIZED

Concept to launch for Tagshop.ai, a GenAI video product

1,000+
REALIZED

Signups in Tagshop.ai's first week, 38+ paying customers in 15 days

$1.2M+
REALIZED

ARR across the Taggbox product portfolio, 30K+ users

60% ↓
REALIZED

Manual moderation effort with AI scoring, UGC Suite (accuracy up 45%)

Outcome labels — REALIZED: shipped and measured in production · MEASURED: result of an eval or test run · DIRECTIONAL: early or small-sample read · OUTPUT: what was built or designed

DomainsAI/GenAI SaaSeCommerce (Shopify, WooCommerce, BigCommerce, Magento, WordPress)MartechEventtechSupply chain & logisticsUS healthcare insurance (EDI, HIPAA)
Claude CodeLovableReplitn8nOllamaLangGraphMCPOnyxOpenAI
ElevenLabsAWS SageMakerHubSpotMixpanelGA4FigmaJIRAPlaywrightShopify
/01

Shipped at work

tagshop.ai as it is today. I led it from concept to launch and left shortly after.

I left Taggbox shortly after launch.

/02

Builds & assignments

Live products I built for take-home assessments or on my own. Each outcome is labeled.

Assignment · Josh Talks · Oct 2026

Akshar Bench

Can AI write India's shop posters?

A blind human evaluation of whether AI image models can put correct Hindi, Gujarati and Punjabi text on a small shop's festive poster. One missing vowel sign turns a sale into a typo, and existing text-to-image boards don't measure it.

20OUTPUT

Fixed prompts

3 + 1OUTPUT

Indian scripts plus English twin prompts

3OUTPUT

Image models compared: GPT Image 1, Gemini 2.5 Flash Image, Gemini 3.1 Flash Image

2OUTPUT

Blind rating rounds

Read the case study →Open live results ↗Methodology ↗
Problem
Image models now make festive posters for small sellers, but nobody measures whether the Indian-script text on them is right.
Bet
Ask the shopkeeper's question first (which poster would you share today?), then audit the text letter by letter.
Result
A live rating app, results leaderboard and methodology, with quality control built in.

Assignment · Aerchain · 2026, then open-sourced

Clearline

Kill the quote spreadsheet.

Buyers lose a week retyping vendor quotes in different formats. Clearline drafts the RFx, reads responses and normalizes one comparison with provenance on every figure.

Multi-format extraction / Agentic analysis / Eval harness / Confidence & provenance / Deployed

34 / 34MEASURED

Eval checks passing

5 × 30OUTPUT

Five vendors, thirty line items

Problem
Buyers lose a week retyping vendor quotes.
Bet
One normalized comparison with provenance.
Result
34 / 34 eval checks passing.MEASURED
Read case study →Live app ↗

Assignment · Entain · 2026

MealsOn

Sell meals, not days.

Meal subscriptions keep failing around a fixed calendar. I replaced it with a meal credit wallet and built a clickable prototype.

Problem reframe / RICE prioritization / 3-phase roadmap / North star metrics / Clickable prototype

6OUTPUT

Connected prototype screens

9OUTPUT

Prioritized initiatives

0OUTPUT

Cost to the user when a day is skipped

Problem
Meal subscriptions keep failing.
Bet
Sell meals, not fixed calendar days.
Result
Six connected prototype screens.OUTPUT
Read case study →Try prototype

Assignment · MYCPE ONE · 2026

Email Verifier

Confidence, not a yes/no.

A bulk email deliverability checker for a marketing team. Instead of a false valid/invalid flag, it scores each address in layers: syntax, MX records, disposable domains and role accounts, so marketing can decide how much risk to send to.

Layered scoring / Syntax / MX / Disposable domains / Role accounts

Problem
Bulk lists mix good, risky and dead addresses, and a binary check hides which is which.
Bet
Layered confidence scoring the team can threshold.
Result
A working internal tool in use behind a login, because it holds real contact data.
Code on GitHub ↗Live app (login required) ↗
/03

Recent builds

Early beta · live

Bravo

A free, open-source AI agent that runs on your own Windows PC. It triages your inbox, browses the web in its own Chrome window, sets reminders and dictates into any app, talk or type, English or Hindi. It asks before it does anything risky, and your voice stays on your PC. Runs on Ollama or any AI key you bring. Early beta, built solo, named after my dog.

Local-firstOpen source (MIT)OllamaVoice + chatApprovals enforced in code
OUTPUT
Visit site GitHub
bravo-agent.lovable.app, the landing page I designed and shipped.
In progress

Event photo and passes platform

A web rebuild study of an event photo-sharing and passes product, built with a design-system-first workflow: recon, design tokens, components, then screens. Scoped to 12 product areas on web, as a product I may sell.

Product teardownDesign systemWeb
OUTPUTNot yet released
Product bet

Parakh

Quality testing for Hindi, Hinglish and other Indian-language voice and chat AI agents. I wrote the discovery kit and PRD: problem, positioning, GTM and metrics.

AI evalsIndian languagesPRD
OUTPUTNot yet released
PRD

AI Shelf

A PRD for an AI product idea for merchants, with discovery, positioning, GTM and metrics.

PRDeCommerce
OUTPUTNot yet released
Live

MyBuddy

A personal AI knowledge assistant built on the Onyx open-source stack, extended with a Google Drive connector and the ability to spin up agents over my own documents. Wired to Ollama so it can run against local models instead of hosted APIs.

Runs locally

Notes AI

A real-time AI meeting assistant built on the open-source Natively codebase. It does the job Natively and Cluely do: live transcription, meeting notes and in-call answers, but runs locally on Ollama models, so the conversation never leaves the machine.

In use

n8n automations

Workflows that remove repetitive work. At Aelum, a sales outreach agent took each rep to 100+ personalized outreaches a day: Google research, Apollo and Sales Navigator for decision-makers, drafted email and LinkedIn messages, and every lead logged in HubSpot. At Taggbox, recurring internal workflows automated as part of the team's AI adoption.

/04

How I work

  1. 01

    01 / Prototype first

    Build it before you spec it.

    A working prototype settles questions a document can only argue about.

  2. 02

    02 / Cost is a product decision

    Latency, inference cost, quality.

    I weigh user value against latency, inference cost and quality, not just whether a feature demos well.

  3. 03

    03 / Surface uncertainty

    Never confidently wrong.

    Confidence scoring, provenance and human review beat a clean interface that quietly guesses.

  4. 04

    04 / AI in the workflow too

    Not just in the product.

    At Taggbox I led AI adoption across teams with agentic coding, QA, PRD tools and n8n, while keeping decisions human.

  5. 05

    05 / Say what's measured

    Realized, measured, or directional.

    Production results, eval results and early reads carry distinct labels so a reviewer knows which kind of evidence they’re looking at.

/05

Experience

  1. Apr 2026 to Aug 2026

    AI Product Manager

    Aelum Consulting

    Led a 7-person team building AI agents and automation for clients and internal use.

  2. Jul 2025 to Apr 2026

    Senior Product Specialist, AI & Computer Vision

    KoiReader Technologies

    Five CV use cases for warehouse and yard operations; camera and edge-server procurement across vendors.

  3. Mar 2023 to Jul 2025

    Product Manager, AI & GenAI SaaS

    Taggbox

    Led a 23-person engineering, AI and QA team. Managed Taggbox, UGC Suite, Taggbox Commerce and Tagshop.ai: strategy, pricing, GTM, launches and customer demos. $1.2M+ ARR; users scaled from 30K to 65K.

Toolkit

AI
LLMs, AI agents, agentic workflows, RAG, vector databases, embeddings, MCP, LangGraph, evals, prompt engineering, open-weight models (Mistral, Gemma) via Ollama, computer vision (YOLO, DETR, PaliGemma), NLP, recommendation systems
Product
0-to-1, strategy and positioning, roadmap and prioritization, PRDs, discovery, A/B testing, OKRs, pricing, GTM
Tools
Mixpanel, GA4, Clarity, Mouseflow, Power BI, Figma, n8n, Claude Code, Replit, Lovable, Windsurf, GitHub Copilot, Playwright, AWS SageMaker, RunPod, JIRA

Education

MBA, IT & Service Management

NMIMS

B.Tech, Computer Science

Global Institute of Technology, Jaipur

Certifications · CSPO · PMI-ACP · PRINCE2

/06

Straight answers

Jitendra Singh · Open to roles

Let’s build the next one.

jitendrashekhawat00@gmail.com
Email+91 85599 91626LinkedInGitHubDownload CV
↑ Top