Product Management Portfolio

I own the why.
I can also ship the what.

I'm Manaswin Pandey, a Senior Product Manager based in Gurugram with 12+ years across energy, strategy consulting, and enterprise life sciences SaaS. Currently at Indegene, leading a 28-person cross-functional team building AI-powered commercial products for global pharma clients.

I'm also serious about AI — not as a talking point, but as something I build with. I've shipped Gen AI into production inside one of the most regulated industries on the planet. And I've taken products from idea to deployed MVP over a weekend using Claude Code.


The Short Version
Most PMs talk about AI. I build with it. I've shipped a Gen AI content pipeline serving 2 enterprise pharma clients across 7 markets — cutting turnaround times by 50%. I've taken a full-stack SaaS from concept to deployed product in a weekend using Claude Code. I know the difference between Sonnet and Haiku, and more importantly, I know when to use which one and why.

But here's the thing: the AI work sits on top of a decade of building across genuinely different contexts. I started as a process engineer at Shell — optimising gas plant uptime with sensor data. Then strategy consulting to an MD at ONGC Videsh, building $50M acquisition business cases at board level. Then product leadership at Indegene, where I've secured investment for three 0→1 products, led teams of 28, and earned two promotions in four years.

AI is the latest chapter. And it's changing how I think about every chapter before it.


Where I Work

Indegene

Senior Product Manager, leading cross-functional teams of 28 — 3 PM direct reports, 25 engineering indirect reports — across the commercial content suite for global life sciences clients.

Commercial Content Suite · Life Sciences B2B SaaS
Gurugram  ·  Mar 2020 – present
Senior PM
Revenue generated
$4M+
SaaS license across 0→1 launches
Content turnaround
−50%
Gen AI pipeline in production
Operational efficiency
+60%
AI/ML content tagging
Requirements rework
−50%
MVP-led validation
  • Gen AI content generation — pharma marketing: Led the build of a production AI pipeline creating derivative marketing assets from pre-approved content. Extraction → semantic mesh → intent-based retrieval → final output. Serving 2 enterprise clients across 7 markets, ~50–60 DAU, contributing ~30–40% of SaaS license revenue. In a regulated industry where most enterprise AI is still a pilot, this is in production.
  • 0→1 product investment: Secured senior leadership buy-in for three new products through market assessment and competitive benchmarking. Translated business cases into $4M+ in SaaS license revenue through go-to-market execution.
  • AI/ML content tagging: Eliminated a manual bottleneck by shipping AI/ML-based tagging. 60% improvement in operational efficiency for content management workflows.
  • Agentic Jira automation: Built a multi-agent system generating structured user stories from high-level inputs — giving junior PMs a requirements scaffold and reducing sprint planning back-and-forth.
  • Customer-led product development: Established product-market fit through structured interviews with domain experts, analysts, and end customers — reducing requirements rework by 50%.
PMO Excellence Award — Commercial Content Platform (May 2025)
Promoted twice in 4 years — top 5% performer
ONGC Videsh
Strategy Consultant to MD
2018 – 2020

Built the business case and renewables integration strategy that secured Executive Committee and Board approval for a $50M Singaporean fuel cell start-up acquisition.

Led portfolio optimisation across $1.5B South American assets — gap analysis with Shell and Qatar Petroleum stakeholders. Reduced cash call processing time by 33%.

Royal Dutch Shell
Senior Operations Engineer
2012 – 2017

Achieved record 98% plant availability through sensor data analysis and control instrumentation optimisation. Key contributor to BG Group Annual Award 2015.

Managed water dumping project resulting in $1M annual increase in production revenue. Mentored 5 operators during crunch periods, maintaining 95%+ KPI performance.


AI Proficiency Timeline

How I got here

This wasn't a single decision. It was a series of deliberate ones — each pulling me deeper into the AI ecosystem and expanding what I could build independently.

2024
Formal foundation: Advanced AI Product Leadership

Completed the Advanced AI Product Leadership Certification on Maven — a practitioner-taught cohort covering AI strategy, model evaluation, and product decision-making in AI-native contexts. Taught by Dr. Marily Nika, Constantinos Neophytou, George Zoto, and Mark Cramer. This was the structured foundation that preceded everything built after it.

AI product strategyModel evaluationMavenCohort-based
Late 2024 / Early 2025
Discovery: AI as a thinking partner

Started with Claude and ChatGPT for everyday PM work: drafting PRDs, sharpening strategy documents, running competitive analysis. This phase was about developing a feel for what these models are genuinely good at, where they fall short, and how to prompt them with enough precision to trust the output.

Prompt engineeringClaudeChatGPTGemini
Mid 2025
First build: no-code AI tools

Built a quiz generation app — Lovable first for speed, Replit when I needed more control. First time going from "using AI to write" to "using AI to build." Discovered what AI-native builders abstract away, and where they hit a ceiling.

LovableReplitNo-code to low-code
September 2025
Levelling up: component-based frontends

Built thewaypoint.ai using 21st.dev's component library. Moved from drag-and-drop into composing real React components with Tailwind CSS. The mental model shifted from "pages" to "systems of reusable parts."

21st.devReactTailwind CSSComponent architecture
Late 2025
Enterprise AI: RAG, cloud infrastructure, production

Shipped the Gen AI content generation pipeline at Indegene — LLMs and vision models for compliant pharma marketing, with semantic retrieval over pre-approved content libraries. First time owning a production AI system in a regulated industry. Human-in-the-loop review was a deliberate architecture decision, not a limitation to overcome.

LLMVision modelsSemantic retrievalRAGGoogle CloudHITL workflow
Early 2026
Full-stack: databases, APIs, caching, deployment

Built Portfolio IQ end-to-end using Claude Code — Supabase, Upstash Redis, Finnhub, FRED, Vercel CI/CD. Went through the full design workflow: Google Stitch for exploration, Figma Pro for production design, Figma + MCP to connect designs directly to Claude-generated React code.

Next.js 14SupabaseClaude APIClaude CodeFigma + MCPVercel
March 2026 to present
Agentic workflows: building tools for teams

Built the Executive Update Copilot — a multi-agent application on Vertex AI giving junior PMs a scaffold for leadership communication. Core principle: AI-assisted judgment, not AI-asserted judgment. Deliberately incomplete outputs with bracket placeholders force the PM to own the judgment calls. Actively used by my direct reports.

Vertex AIGeminiFastAPIMulti-agentGoogle Cloud Run

What I've Been Building

Things I built because I needed to know if I could

Each project started with a question that mattered to me as a PM. The builds were the way to find the answer.

Gen AI Content Generation

The question: Enterprise AI adoption in regulated industries moves slowly — for good reason. Could you build something that genuinely works inside those constraints, not around them?

Shipped · Production · Enterprise

What I learned: The hardest product decisions weren't about the AI model. They were about information architecture — structuring the content library so retrieval was precise, not approximate. And about scope: knowing which parts of the workflow AI could own versus which needed a human in the seat. Human-in-the-loop isn't a limitation to overcome — it's the right call for where enterprise AI adoption actually is today.

Enterprise clients
2 clients
Global pharma
Markets in production
7 markets
~50–60 DAU
Revenue contribution
30–40%
Of $4M+ SaaS portfolio
Turnaround improvement
−50%
Client content projects
LLMVision modelsSemantic retrievalRAGHITL workflowLife sciences
Executive Update Copilot

The question: Junior PMs often know what happened — but struggle to frame it for leadership. Could AI give them a scaffold to think through before anything goes upward?

Shipped · In Use

What I learned: The hardest design decision wasn't technical — it was deciding what the AI shouldn't do. An AI that writes the update for you creates passive PMs. An AI that asks "what's the business impact?" and leaves space for the answer creates sharper ones. That line between assistance and abdication is a product judgment call. I made it deliberately.

Architecture
Multi-agent
Workflow, crisis, exec comms agents
Design principle
HITL by design
Brackets force judgment, not abdication
Vertex AIGemini 2.0 FlashFastAPIGoogle Cloud RunMulti-agent
Portfolio IQ

The question: Can a product manager with no engineering background own the full technical lifecycle of an AI product — from database design to deployment?

Live in Production

What I learned: Writing a thorough PRD before touching any code is what makes AI-assisted development fast. The skill isn't coding — it's specification, architecture decisions, and understanding cost trade-offs between models. That clarity is the same skill that makes cross-functional programmes work at scale.

Build time
A weekend
Focused, end-to-end
Project-specific cost
$5
Claude API; rest is free tier
AI strategy
Single vendor
Anthropic: Sonnet + Haiku
Purpose
PM → full-stack
Non-coder building end-to-end
User → Cloudflare → Vercel → Next.js 14 → Serverless API routes → Upstash Redis → Claude Sonnet / Haiku → Supabase (Postgres + Auth + Storage) + Finnhub + FRED

Frontend in Next.js 14 with Tailwind CSS and shadcn/ui. Backend via serverless API routes — no dedicated server. Redis caching reduces Claude API calls by ~70% for repeated market queries. Supabase Row Level Security means the database enforces permissions, not application code.

  • Google Stitch: Fast initial generation of design system and core screens. Great for exploration, limited for precision.
  • Figma: Production-grade design system — proper component library, variants, auto-layout, design tokens, responsive breakpoints.
  • Figma + MCP: Connected Figma to Claude via Model Context Protocol. Claude reads Figma specs directly and generates matching React code. The handoff isn't a PDF anymore — it's a conversation.
ServiceWhat it doesCost
SupabaseDatabase, auth, file storage$0
Upstash RedisCaching layer$0
FinnhubStock data + news$0
FREDMacro indicators$0
VercelHosting + domain$0
CloudflareCDN + security$0
GitHubCode + CI/CD$0
Google StitchInitial design$0
Claude APISignal generation + parsing$5 credit
Claude Pro (incl. Claude Code)*AI IDE + dev workflow$20/mo
Figma Pro*Design system + MCP bridge$15/mo
Project-specific spend$5

* Claude Pro and Figma Pro used daily for work — not costs specific to this build.

  • Why single vendor? Claude Sonnet handles vision just as well. One vendor = one API key, one bill, one set of docs. Saved $5–10/month with zero capability loss.
  • Why Supabase? Postgres + auth + storage in one free tier. Row Level Security means the database enforces permissions, not my code.
  • Why near-zero cost on purpose? No financial risk means more experiments, more learning — and a real understanding of AI product economics.
  • Why Claude Code? 10 sequential prompts. Each builds on the last. The PRD written before touching code is what made this fast.
01Design tokens: colours, type, spacing
02TypeScript interfaces for all domain objects
03Supabase schema, 7 tables, RLS policies
04Market data service + Redis caching layer
05Claude signal engine + portfolio parsers
069 API routes (signals, portfolio, market, watchlist)
07Auth + Google OAuth
084-step onboarding wizard
09Dashboard, signal cards, detail drawer
10Polish, CI/CD, deploy to Vercel
Next.js 14SupabaseClaude APIClaude CodeFigma + MCPVercelUpstash Redis
Waypoint.ai

The question: What's the difference between assembling an app from AI-generated blocks and composing a product with real design intent?

Shipped

What I learned: Composition is a design skill, not just a technical one. Choosing which components to use, how they relate to each other, and where to break from the library's defaults requires the same product judgment as deciding what goes on a roadmap.

Component source
21st.dev
AI-native UI primitives
Key takeaway
Architecture > assembly
Composition is a skill
21st.devReactTailwind CSS
Quillion — Quiz Generation App

The question: Can AI tools take a product idea from zero to functional without writing code?

Shipped

What I learned: AI app builders are excellent for validation and poor for iteration. The v1 comes fast, but the moment you need to deviate from what the tool assumed, you're fighting it. I now evaluate AI tools the same way I evaluate any product: by their constraints, not their demos.

Build time
~5 hours
Across two platforms
Key takeaway
AI generates; PMs curate
Lovable = speed, Replit = control
LovableReplitJavaScript

Putting It Together

Where strategy meets building

Three industries. Three roles. One consistent thread: the ability to read a complex system, identify the highest-leverage problem, and move toward a solution that scales.

A decade of building across industries and stages.

$4M+ SaaS revenue through 0→1 product launches. 50% improvement in client content turnaround. 60% operational efficiency gain. Two promotions in four years. PMO Excellence Award 2025. These came from knowing which problems to solve and in what order.

Operator instincts inside a product leadership role.

I started as an engineer — optimising real systems under real constraints. Before product management, I built a business case that moved $50M at board level. That background shapes how I think about product decisions: with a strong bias toward what's actually true about the system, not what's convenient to assume.

AI as a genuine capability, not a talking point.

Five projects built, four in production. A Gen AI pipeline serving enterprise pharma clients globally. A full-stack SaaS deployed over a weekend. An agentic tool actively used by my direct reports. The progression from using AI to building systems that run on AI creates a different kind of product intuition — one that matters in roadmap conversations, architecture reviews, and decisions about where AI creates real value versus where it's just a feature checkbox.

Velocity that comes from clarity.

Portfolio IQ went from concept to deployed product in a weekend. The Gen AI pipeline went from a scoped need to production across 7 markets. The Executive Copilot went from a team friction I noticed to a tool my junior PMs use every day. That comes from clear specs, right tools, and knowing which decisions to make now versus defer. In teams, it shows up as crisp PRDs, unambiguous success criteria, and fewer realignment cycles.


Say hello.

I'm always open to conversations about product strategy, Gen AI in regulated industries, or building at the intersection of product and engineering. If something here resonated, I'd welcome a conversation.