Open to AI product roles

Hey there,

I'm Jyotishman

A product manager with 7+ years in B2B SaaS and HR-tech, building products around where AI actually fits. Sole PM for Vantage Pulse at Vantage Circle, shipping LLM-powered features end-to-end and prototyping live in the production codebase with Claude Code.

See the case studies →
Jyotishman Das, Product Manager
0→1
Rebuilt a dormant add-on into a standalone paid product
7+
Years of product impact
AI-native toolkit
Claude CodeCodexCursorFigmaJiraHotjarPendo
How I evaluate an AI product

Good AI products aren't features bolted on. They're workflows redesigned.

Before I scope anything with an "AI" label, every idea has to answer five questions. It's the same lens I use to pressure-test my own roadmap: diagnosis before intervention, and honesty about the tradeoffs.

01 / WORKFLOW

What is the job today?

Map the real workflow before AI: who does what, in what order, with which tools. You can't redesign a process you haven't watched.

02 / FRICTION

Where does it actually break?

Name the specific friction: volume, latency, inconsistency, cognitive load. Vague pain gets vague solutions.

03 / CAPABILITY

Which AI capability targets it?

Match a concrete capability (classification, RAG, summarization, generation) to a concrete friction point. Not "add an LLM."

04 / IMPACT

What moves, and how do I know?

Tie it to a measurable business outcome with the reasoning shown, not a number pulled from the air.

05 / PRODUCT THINKING

What are the risks, and who stays in the loop?

The part most demos skip: failure modes, confidence thresholds, human-in-the-loop approval, fallback triggers, and the prerequisites that have to be true first. This is where trust is won or lost.

Selected work

Three AI workflow redesigns, end-to-end.

Each is real product work I owned at Vantage Pulse, structured the way I think about AI: the workflow before, the friction, the intervention, the measurable impact, and the product decisions that made it safe to ship.

◆ NLP · Classification · Assisted Generation

AI Feedback Intelligence for Employee Engagement

Turning thousands of open-text survey responses into themes HR can act on, and replies HR can send with one review.

Role: Sole PM Authored the classification prompts Defined the theme taxonomy Eng + Data partnership
Impact
~70%
Estimated synthesis-time reduction per pulse cycle: from manually reading & tagging every comment to reviewing pre-clustered themes. Basis: a 500-response cycle at ~30s/comment ≈ 4hrs of reading, collapsed to ~1hr of review.
+25%
Weekly active engagement uplift the feature line contributed to (Best Team Award).
+15%
Enterprise client retention improvement across accounts via tighter feedback-to-action loops.
01 · Workflow before

Manual reading, manual replies

Each pulse cycle, HR admins opened a spreadsheet of open-text comments, read every one, mentally grouped themes, pasted highlights into a deck, and hand-wrote individual replies to employees.

02 · The friction

Volume broke the loop

  • Thousands of comments per cycle across 60+ clients
  • Inconsistent theming between admins
  • Reply fatigue → many employees never heard back
  • Days between feedback and action
03 · AI intervention

Classify, cluster, assist

  • Sentiment + theme classification into a defined taxonomy
  • LLM summarization of each cluster
  • AI Reply Assistant drafts context-aware responses for admins
sentiment_theme_classifier.prompt
# Classify one employee comment. Output STRICT JSON only.
ROLE: You classify anonymous employee feedback.
TAXONOMY: [Compensation, Management, Growth,
  Workload, Culture, Tools, Recognition, Other]

RULES
- sentiment ∈ {positive, neutral, negative}
- theme MUST come from TAXONOMY
- If confidence < 0.6 → theme = "Other",
  set "needs_human_review": true
- NEVER invent themes. NEVER quote the
  employee by name. NEVER guess identity.

OUTPUT
{ "sentiment": "...", "theme": "...",
  "confidence": 0.0-1.0,
  "needs_human_review": false }
Product thinking
Human in the loop

The Reply Assistant drafts. HR admins review and approve every message. Nothing is ever auto-sent to an employee.

Fallback trigger

Confidence below threshold routes the comment to a manual-review queue instead of a low-quality auto-tag.

Guardrail

Anonymity is preserved end-to-end; the model never receives or infers employee identity. Sentiment is advisory, never a performance signal.

◆ RAG · Knowledge Base · LLM

AI HR Knowledge Assistant

A grounded chatbot that answers repetitive HR questions from company documentation, and knows when to escalate.

Role: Product owner Defined requirements Eng + Data science build Launch owner
Impact
~60%
Estimated deflection of repetitive policy/benefits/process questions away from HR inboxes. Basis: the long tail of HR queries is highly repetitive and answerable from existing docs.
24/7
Instant answers vs. waiting on a human during business hours.
1 source
One grounded knowledge base replacing scattered docs and inconsistent verbal answers.
01 · Workflow before

HR as a human FAQ

Employees pinged HR over email and chat with the same policy, leave, and benefits questions. HR answered manually, from memory or by hunting through scattered documents.

02 · The friction

A repetitive bottleneck

  • Same questions, over and over
  • HR became a blocking dependency
  • Answers varied by who replied
  • Employees waited hours or days
03 · AI intervention

Grounded retrieval + LLM

  • Company docs indexed as a knowledge base
  • LLM answers only from retrieved context
  • Confident, cited answers, or a clean handoff to a human
grounded_answer.prompt
# Answer HR questions from retrieved context ONLY.
CONTEXT: {{retrieved_policy_chunks}}
QUESTION: {{employee_question}}

RULES
- Answer ONLY from CONTEXT. Cite the source doc.
- If CONTEXT does not contain the answer →
  do NOT guess. Reply:
  "I don't have that in policy docs,
   connecting you to an HR partner."
- Sensitive topics (grievance, health,
  termination) → always escalate to a human.

TONE: clear, warm, concise. No legal advice.
Product thinking
No hallucinated policy

The model is constrained to retrieved documents and cites its source. If the answer isn't in the docs, it says so. It never improvises policy.

Escalation by design

Sensitive or confidential topics bypass the bot entirely and route to a human HR partner.

Prerequisite

Only as good as the knowledge base, so doc freshness and coverage were tracked as a first-class product metric, not an afterthought.

◆ AI-Assisted Development · Process Redesign

AI-Assisted Product Development

Prototyping live in the production codebase with Claude Code & Codex, collapsing design-to-dev from weeks to days.

Role: PM, hands-on Production Angular codebase Paired with frontend eng AI Adoption Pioneer award
Impact
2–3 days
From weeks to days: feature redesigns prototyped and deployed live instead of cycling through static mockups.
−20%
Time-to-market reduction on the delivery squad through tighter iteration loops.
Team-wide
The workflow I introduced is now used across the Vantage Circle product team.
01 · Workflow before

Mockups over the wall

Ideas became static Figma mockups, handed to engineering, translated into code, reviewed, revised. Every loop cost days, and PMs couldn't feel an idea in the real UI until late.

02 · The friction

Slow, lossy iteration

  • Weeks between idea and testable UI
  • Translation loss from mock to build
  • Expensive to explore alternatives
  • Feedback arrived too late to cheaply change
03 · AI intervention

PM prototypes in real code

  • Claude Code / Codex directly in the Angular codebase
  • Working redesigns, not throwaway mocks
  • Paired with frontend eng to ship in 2–3 days
prototyping_workflow.md
# How AI prototyping fits, with guardrails
1. SCOPE   PM drafts the change on a feature
          branch, never on main.
2. BUILD   Claude Code implements against the
          real component + design system.
3. REVIEW  Frontend engineer reviews every diff
          before merge. No PM-authored code
          ships unreviewed.
4. QA      Standard QA + Definition of Done
          still gate the release.
5. SCALE   Document the pattern → hand it to
          the wider product team.
Product thinking
Where it fits (and doesn't)

AI prototyping is for exploring and de-risking UI, not for shipping unreviewed PM code to production. The gate stays with engineering.

Risk managed

Feature branches + mandatory eng review + unchanged QA guardrails prevent "PM overreach" while keeping the speed.

Systems view

The real win wasn't my speed. It was making the practice repeatable and adopting it across the team.

Like how I think? Let's talk it through.

20 minutes on your AI product, your roadmap, or a role you're hiring for. No deck, just product talk.

Book an intro call →
Beyond my backlog

Products I love, torn down.

I study products the way I build them: what is the job, where is the friction, and why does the design win. A few favorites from my teardown notebook. Full teardowns available on request.

Linear

Project management

What it nails: speed as a feature. Keyboard-first, sub-100ms interactions, and opinionated defaults that remove decisions instead of adding settings.

What I'd push: leadership-facing insights lag the IC experience; roll-up reporting still sends execs back to spreadsheets.

Perplexity

AI answer engine

What it nails: citations as the product. Grounding every answer in visible sources turns AI skepticism into trust, the same principle I applied to my HR knowledge assistant.

What I'd push: session memory. Research threads reset too easily, losing the context a returning user has already built.

Duolingo

Consumer learning

What it nails: habit mechanics. Streaks, leagues, and loss aversion turn motivation into a system, the best engagement loop design in consumer software.

What I'd push: the ceiling. Past intermediate level, gamification outpaces pedagogy; retention stays but proficiency plateaus.

Stripe

Payments infrastructure

What it nails: documentation as the product. Seven lines of code to accept payments; the developer experience is the moat, not the rails underneath.

What I'd push: surface sprawl. As the product suite grows, the dashboard asks users to hold more of Stripe's org chart in their heads.

Notion

Workspace tools

What it nails: the block as a primitive. One composable unit powers docs, databases, and wikis, and templates turned users into the growth engine.

What I'd push: the blank-canvas problem. Infinite flexibility is a tax on new users; day-one guidance still leans too hard on templates.

Figma

Design collaboration

What it nails: multiplayer as the moat. Browser-first killed install friction, and comments pulled PMs and engineers into design instead of exporting it to them.

What I'd push: file governance at org scale. Finding the source of truth across teams, pages, and branches still relies on tribal knowledge.

Experience

Where the reps came from.

Jan 2023 – PresentVantage Circle

Product Manager, Vantage Pulse

B2B SaaS employee engagement & automation platform · Sole PM, reporting to the co-founder

0→1 ownership

Rebuilt a dormant, unresearched add-on into a standalone product with active paid adoption across 60+ enterprise clients.

Shipped the suite

Automated lifecycle surveys, kiosk/QR access for employees without email, and ad-hoc pulses; replaced a fully manual cadence.

LLM features live

Authored the sentiment-classification prompts and theme taxonomy; shipped an AI Reply Assistant and an HR chatbot with eng and data teams.

2–3 day ships

Pioneered AI prototyping with Claude Code and Codex in the production Angular codebase; design-to-dev went from weeks to days.

Sole GTM bridge

The product-side link across every function: partnering with Engineering, Design, and QA to build, briefing Marketing ahead of feature launches, arming Sales/SDRs for outbound and expansion, syncing with Customer Success on adoption and renewal signals, and tracking ARR impact with Finance and Leadership.

Hotjar funnel analysisVoice of CustomerQuarterly exec reviewsAgile squad lead
Jan 2019 – Dec 2022Nihilent

Associate Consultant, Business Analysis & Product Ownership

Enterprise digital transformation engagements

Product Owner proxy

Owned backlog and sprint delivery for a Contract Management System and a Portfolio Management Service for one of India's largest asset managers.

5+ programs

Concurrent client engagements across India, South Africa, and the US in a matrixed global org, with a perfect on-time record.

Emotion-AI bridge

First business-analyst bridge on EmoScape: translated business goals into ML requirements for a system processing 75+ behavioral signals.

100+ trained

Led Design Thinking workshops with banking clients and trained 100+ employees; A/B tests and SQL analysis drove backlog priorities.

Design Thinking facilitationA/B testingSQL analysisOffshore delivery
Toolkit

What I bring to a product team.

Product

Product StrategyRoadmap Ownership0→1 Delivery PRD / BRD / FSDRICE · MoSCoW · WSJFAgile / Scrum Voice of CustomerDesign ThinkingGo-to-Market

AI & Building

LLM Feature ScopingPrompt EngineeringRAG Claude CodeCodexCursor NLP / SentimentAgentic WorkflowsPostman

Analytics

SQLA/B TestingCohort Analysis Funnel OptimizationKPI / OKR DefinitionHotjar PendoGoogle AnalyticsMixpanel · Amplitude

Education

MBA / PGDM, Marketing & Business Analytics, Balaji Institute of Telecom & Management, Pune (2017–2019)

B.E., Electronics & Telecommunication Engineering, Gauhati University (2011–2015)

Certifications & Recognition

  • Certified Scrum Product Owner (CSPO), Scrum Alliance
  • Product Analytics Certification, Pendo
  • Enterprise Design Thinking Practitioner, IBM
  • Awards at Vantage Circle: Star Performer, AI Adoption Pioneer, Best Team, Standout Performer
Social proof

What colleagues say.

What stands out most about JD is his ability to approach challenges from perspectives that others often miss. He combines deep analytical thinking with genuine empathy... His ability to independently figure things out, coupled with his strong work ethic and genuine care for people, makes him an invaluable asset to any team.
DP
Diksha Pandey
Client Care Advocate · worked with JD on the same team
Quiet, thoughtful, and deeply observant... when he spoke, it was always with clarity and purpose. He stood up for the team when it mattered, with calm conviction and a clear sense of what was right. I'd gladly recommend him to any product team looking for someone who combines maturity, integrity, and a genuine team-first mindset.
AJ
Anand Jeyaraman
Design Thinker & TEDx Speaker · JD's mentor at Nihilent
Let's talk

Building something with AI at its core?

I'm most useful where a real workflow needs rethinking, not where AI is a checkbox. If that's the kind of product you're building, let's talk.

📅 Book a 20-min intro call ✉ Email me in · LinkedIn