Nilank
Udupi
I build AI products that ship and stick.
PM across GenAI tooling, growth, and founding. I own roadmaps end-to-end - from prompt engineering and API cost models to production dashboards - without losing sight of the user who has to live with the result.
I started as a founder. Built eGalleria - one of India's first digital exhibition platforms for artists - from scratch to 100,000 monthly users. Not through funding or paid acquisition, but through relentless prioritisation and distribution that actually worked. Before that, co-founded JustNCase and grew revenue ~95% month-on-month. Both taught me the same lesson: the only thing that matters is whether users show up and stay.
I crossed into product consulting and brought the founder lens with me. At GetDefault I ran four client engagements in parallel. At Zeiss I took a process that required an optical expert and redesigned it into a guided staff tool with competency checkpoints - consultation time down 36%. At WizKidsCarnival I rebuilt the entire participant journey, deployed a Gemini-powered AI pipeline that cut content costs by 87%, and built a Claude API analytics dashboard that eliminated analyst dependency for campaign decisions.
Most recently at BoutiqBiz, I owned three production GenAI tools on Gemini 2.5 Flash and Veo 3.1. I wrote the prompt templates, built the API cost models, caught a 36% pricing underestimate before it hit decisions, and shipped the dashboard. Production bugs dropped 85%. The founder stopped firefighting and started building.
I go deep on technical decisions without losing the user. I don't hand off and wait - I sit in the spec, model the costs, write the prompts, and come back with something to show. That's the edge.
How I work with AI
Not "I use AI tools." I own the layer between the model and the person using the output. Here's what that looks like in practice.
Where I've shipped
- Rebuilt end-to-end participant journey from discovery to on-stage performance; active users +80%, platform efficiency +73%
- Deployed Gemini-powered AI contest pipeline using structured policy doc as context layer; content costs Rs.40K to Rs.5K/month (-87%), 20 contests/month maintained
- Built Claude API-powered internal analytics dashboard (4-tab React app): Meta spend tracking, AI-generated per-day commentary, Ask AI freeform query - zero analyst dependency for campaign decisions
- Designed and shipped Showcases platform - live event site for Mumbai and Pune performing arts events, handling real registrations in production
- Drove persona-based free-to-paid conversion redesign; revenue +30%
- Owned roadmap for three production GenAI tools - Photoshoot Gen, Style Gen, Video Gen - on Gemini 2.5 Flash and Veo 3.1
- Wrote v2 prompt templates with separation of concerns: reference images for visual description, Veo prompts for motion only - eliminated a full class of hallucination bugs
- Built API cost models; caught a ~36% underestimate (Vertex AI vs Gemini API rates) before it influenced commercial decisions
- Shipped two-screen AI dashboard, scoped all briefs in Jira. Production bugs -85%; founder firefighting time -70%
- Zeiss: Turned an expert-dependent optical consultation process into a guided staff flow with pass/fail gates; consultation time -36%
- WizKidsCarnival: Led platform from zero to 85,000 MAUs; built white-label onboarding framework scaled to major brand partners
- Elevation Capital: Restructured internal knowledge base into role-based directory; operational workload -54% via automation
- RightLife: PM'd Sonde and Anura ML API integration for health diagnostics
- Built from zero to 100,000 MAUs; digitised exhibition submission and curation end to end
- Owned branding, growth, and product design; 80% MoM growth at peak on cost-efficient distribution
- Built a customised smartphone case brand from scratch; ~95% MoM revenue growth through scrappy branding and supply chain ownership
Selected work
Four projects with the decisions that actually drove the outcome - not just the metrics.
The problem: BoutiqBiz had three GenAI tools in development and no PM to tie them together. Each tool had separate engineering assumptions, no shared prompt discipline, and an API cost model built against the wrong pricing tier.
The first fix was the cost model. The team had priced against Gemini API rates; the actual invoices would come through Vertex AI. I caught the ~36% gap during a pre-launch pricing review and rebuilt the model before any commercial decision got anchored to the wrong number.
On prompts, I wrote v2 production templates with an explicit separation of concerns: reference images handle visual description only, Veo prompts handle motion and camera behaviour only. This eliminated the class of bugs where the model inferred motion from a static product swatch.
Ran weekly engineering reviews, scoped all briefs in Jira, shipped a two-screen AI dashboard. Production bugs dropped 85%. The founder stopped triaging fires and started working on growth.
The original WKC participant flow was fragmented - discovery on Instagram, registration in a Google Form, confirmations over WhatsApp, logistics in a spreadsheet. Nothing talked to anything else. Enrolment dropped at every step and nobody knew where.
I mapped the full journey from first touchpoint to a child walking onto the stage. Defined enrolment rate and completion milestones as the primary KPIs. Rebuilt the flow into a single landing page, inline registration form, automated confirmation sequences, and an admin layer the team could manage without help.
Designed and shipped showcases.wizkidscarnival.com - live in production, serving the Mumbai and Pune events. The site handles registrations, communicates event details, and converts parent interest at a measurably higher rate than the old fragmented flow.
In parallel I ran user interviews across three parent personas and rebuilt the free-to-paid conversion journeys around their specific hesitations. Revenue grew 30% in the first cycle. Active users are up 80% - and unlike the old numbers, these are tracked, not estimated.
WKC was spending Rs.40,000 per month on contest content - freelancer time, coordinator review cycles, founder edits. Output was 20 contests per month but quality was inconsistent and the process couldn't scale.
I applied instructional design principles before touching any AI tooling. First I defined what a good contest looks like: age-group rules, format constraints, scoring rubric structure, quality standards. This became a structured policy document - not a prompt, but a codified decision framework.
That policy doc became the context layer fed to Gemini at inference time. The model generates contest briefs, rubrics, and scoring criteria against the constraints in the doc. Because the constraints are explicit, the outputs are consistent without extensive review.
Monthly spend dropped from Rs.40K to Rs.5K - an 87% reduction. We maintained 20 contests per month. The remaining cost is one hour of review per week, not a team of freelancers. The policy doc also made it trivial to add new contest formats: update the doc, the AI adapts.
The team was tracking Meta ad spend, quiz completions, and enrolment in three separate sheets. Correlating them meant manual copy-paste every Monday. Campaign decisions were made on week-old data.
I built a Claude-powered analytics dashboard to collapse this into a single view. No database - a four-tab React app where you upload a telemetry CSV and the dashboard does the rest: parses it, computes derived metrics (CPC, CPL, CTR), and routes the full dataset to Claude for AI-generated per-day commentary.
The key engineering decision was schema fingerprinting: Claude's CSV mapping runs once per schema structure and gets cached in localStorage. Subsequent uploads of the same format skip the AI call entirely - keeping API costs flat without degrading the experience.
The Ask AI tab lets any team member query funnel performance in plain English. No SQL, no analyst, no two-day wait. Decision-making time on campaign performance went from days to minutes.
From the people I've worked with
Nilank's dedication and ingenuity consistently stood out across every project we worked on together. He excels at streamlining processes, enhancing products, and ensuring seamless communication among all stakeholders. He has a knack for tackling complex challenges with creative, out-of-the-box solutions that go beyond conventional approaches.
Nilank is just exceptional at managing all kinds of products. His attention to detail and dedication are top-notch, and he's a great communicator, making collaboration effortless. His contributions to any product he touches are invaluable - I have no doubt he'll continue to excel.
Toolkit
If you're working on a hard AI product problem - or looking for someone who can own it end-to-end - I'd like to hear about it.