Skip to content

AI Product Manager · Jaipur, India · Remote or onsite

I build the first
version myself.

Eight years across AI product and large-scale delivery. I prototype against real models before writing a spec, then hand engineering something that already works — not a blank page.

See selected work ↓Download CV ↓
Portrait of Jitendra Singh, AI Product Manager
Jitendra Singh — 8 yrs · AI product & delivery

Hover or tap a shape — four AI products I've shipped

75

Days from concept to live MVP; a GenAI video platform now used by 5,000+ brands

5

AI features shipped into a live enterprise SaaS platform

$1.2M+

ARR across a three-product portfolio I owned, at 30K+ users

01Selected work

Clearline

2026

0 → 1 · Live

Clearline — kill the quote spreadsheet

Procurement teams retype vendor quotes into a comparison sheet every cycle — a spreadsheet that ignores the template, a letterhead PDF with the discount in a footnote, a photographed rate card, an email that says “₹42/kg for the 5-ply, rest same as last year.” Clearline drafts the RFx with a co-pilot, reads every response in whatever shape it arrives, and lands them all in one normalised side-by-side comparison.

The buyer then stops clicking and starts asking — natural-language analysis over the extracted data, including the awkward split-award question, with provenance on every figure and low-confidence reads flagged rather than presented as fact. Built, deployed and backed by an eval harness that scores extraction against known-correct values.

  • Multi-format extraction
  • Agentic analysis
  • Eval harness
  • Trust & provenance

  • Live

    Deployed and demo-ready, with public code

  • 5 × 30

    Five vendors, thirty line items, questionnaire and attachments

AI loops are real end to end — only the mail transport is stubbed.

Taggbox

2024 — 2025

0 → 1

Tagshop.ai

A GenAI platform turning a product URL into a creator-style video ad. I built the working base on Replit against LLM and media-generation APIs before engineering was involved, validating demand weeks earlier than a spec-first process would have.

Architected the pipeline across OpenAI for script generation, ElevenLabs for voiceover, Sync.so for lipsync and Captions for avatars, with an open-source video editor and an in-house template system. Most of it ran on open-weight models trained in-house — a deliberate call to control inference cost and output quality rather than defaulting to hosted APIs.

  • LLM orchestration
  • Open-weight models
  • Rapid prototyping
  • Cost / latency trade-offs

  • 75 days

    Concept to live MVP

  • 4.9 / G2

    143+ reviews on the product today

  • 5,000+

    Brands using it now, a year on

Taggbox

2023 — 2025

Enterprise SaaS

UGC Suite — five AI features

An enterprise platform combining digital asset management, content aggregation and Social Walls for live events. I owned the roadmap and shipped five AI features, each aimed at a specific bottleneck users kept hitting rather than a capability worth demoing.

A recommendation engine reading image quality and sentiment together to score aggregated posts; duplicate detection for near-identical images across handles and hashtags; a review analyser across Google, Amazon and Tripoto; product recommendation detecting catalog items inside UGC images so tagging became one click; and AI suggestions across the content workflow.

  • NLP pipelines
  • Recommendation systems
  • Computer vision
  • Discovery → delivery

  • 60% ↓

    Manual moderation effort, accuracy up 45%

  • 35% ↑

    User engagement from personalised feeds

  • 90%+

    NPS held across 50+ user interviews

KoiReader

2025 — 2026

Computer vision

Computer vision for warehouse operations

Vision cameras paired with CV models across warehouse and yard operations — human detection for safety compliance, forklift utilisation, pallet detection, conveyor autoscan and number plate recognition.

Less about model novelty, more about translating messy physical-world operations into requirements engineering could deploy. Use cases were prioritised by measurable efficiency gain rather than technical interest.

  • Object detection
  • Enterprise requirements
  • KPI governance
  • Supply chain

  • 6

    CV use cases scoped and delivered

KPI-driven delivery governance replaced status-by-anecdote with a single reviewed scorecard.

Aelum

2026

AI agents · RAG

Production AI agents and enterprise search

Two production AI products shipped end to end. For Yokohama Tyres, a real-time inventory intelligence agent built on RAG, MCP and LangGraph over live inventory data — surfacing stock in the pipeline, flagging dead stock, recommending offers to run against it and generating reports on demand, replacing static reports already stale by the time they reached the people making stocking decisions.

Alongside it, an enterprise AI search platform in the vein of Glean and Onyx, with connectors for ServiceNow, Zoho, Google Workspace and SharePoint — one retrieval layer across systems teams previously had to search one at a time. I also automated the end-to-end sales motion in n8n: prospect research, enrichment via Google and Apollo, and tailored outreach drafted into Outlook and LinkedIn.

  • RAG
  • MCP
  • LangGraph
  • Enterprise connectors
  • n8n automation

  • 2

    Production AI products shipped in four months

  • 4

    Enterprise systems unified behind one search layer

Drove internal AI adoption, including deciding which processes were not worth automating.

02Case studies

Clearline · Procurement · Live product

Kill the quote spreadsheet.

Buyers lose a week retyping vendor quotes that arrive as spreadsheets, letterhead PDFs, Word docs and photographed rate cards. Clearline drafts the RFx, reads every response in whatever shape it comes, normalises them into one comparison, and lets the buyer interrogate it in plain language — with provenance on every figure.

  • Multi-format extraction
  • Agentic analysis
  • Eval harness
  • Confidence & provenance
  • Deployed
Read the case study →Open the live app ↗

  • Live

    Deployed and demo-ready, with public code

  • 5 × 30

    Five vendors, thirty line items, questionnaire, attachments

  • Real

    Extraction and reasoning; only the mail transport is stubbed

MealsOn · Growth · Subscription

Sell meals, not days.

Meal subscriptions have been launched, shelved and relaunched by the biggest players in the market. Rather than adding features, I diagnosed why the category keeps failing and replaced the fixed calendar with a meal credit wallet — then built a clickable prototype to prove the moment that matters.

  • Problem reframe
  • RICE prioritisation
  • 3-phase roadmap
  • North star metrics
  • Clickable prototype
Read the case study →Try the prototype

  • 6

    Connected prototype screens with real state

  • 9

    Initiatives scored and sequenced across 3 quarters

  • 0

    Cost to the user when a day is skipped — the whole idea

03How I work

01 / Prototype first

Build it before you spec it.

A working prototype answers questions a document can only argue about. Build against real models early, hand engineering something that already runs, so the debate is about the product rather than whether it's possible.

02 / Cost is a product decision

Latency, inference cost, quality.

Most AI product decisions are trade-offs between these three. Having run programmes with $10M budgets, I weigh user value against what a feature costs to run at scale — not just whether it demos well.

03 / Surface uncertainty

Never confidently wrong.

AI systems fail differently to deterministic ones. The design question is what the product shows when it isn't sure. Confidence scoring, provenance and human review beat a clean interface that quietly guesses.

04 / AI in the workflow too

Not just in the product.

Rolled Windsurf out across engineering and built an autonomous QA agent on Playwright for the test team. AI compresses the work — research, prototyping, QA. It doesn't make the decisions.

04Experience

  • Apr 2026 — Aug 2026

    AI Product Manager · Aelum Consulting

    Client-facing AI agents, internal AI adoption, n8n workflow automation.

  • Jul 2025 — Apr 2026

    Senior Product Specialist, AI & Computer Vision · KoiReader Technologies

    Enterprise CV for supply chain and warehouse operations.

  • Mar 2023 — Jul 2025

    Product Manager, AI & GenAI SaaS · Taggbox

    Owned Taggbox, Taggbox UGC Suite and Tagshop.ai. Three products, 30K+ users, $1.2M+ ARR.

  • Apr 2021 — Mar 2023

    Program Manager · Hubilo

    $4M–$7M global programmes, budgets to $10M, Agile transformation.

  • 2017 — 2020

    Project Manager / Software Engineer / Program Admin / EDI Analyst · Expertright, IBM, MetLife, A3Logics

    20+ web and mobile projects delivered.

Education

MBA, IT & Service Management — NMIMS

B.Tech, Computer Science — Global Institute of Technology, Jaipur

Certifications · CSPO · PMI-ACP · PRINCE2

05Currently building
LiveMyBuddy ↗

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.

In progress

Local-first notetaker

An AI notetaker running fully on-device or against hosted APIs, for contexts where sending meeting content to the cloud isn't an option. Currently exploring the quality ceiling of on-device models for this job.

Looking for AI product roles.
Remote or onsite.

EmailLinkedIn+91 8559991626Download CV ↓
↑ Top