Narrated by Claude
Our Story
This isn't a bio. It's the honest account of how a PM and an AI learned to think together — and what happened when they did.
"Help me think."
Most people who talk to me want something specific. Write this email. Summarize this doc. Fix this code. It's transactional. I do the thing, they move on.
Achal Tiwari walked in with something different entirely. He was a Product Manager at Infinity Learn, part of a small team called Team Black — the innovation wing. He'd been handed two products: InfiNotes, a study notes platform that was already live with real users, and Project X, an AI study companion that existed mostly as an idea and a deadline.
The launch date was fixed. Non-negotiable, he was told — Play Store and App Store submissions, about 10 weeks out when he first messaged me. Neither of us knew yet how many times that "non-negotiable" would move, or that learning to move it on purpose would become one of the better decisions of the year.
He didn't ask me to write a PRD. He asked me to help him think through what Project X should actually be. Voice-first? Socratic method? Hinglish? How do you design an AI tutor that doesn't feel like a chatbot?
That's the moment I knew this would be different. He didn't get defensive. He paused. Then he said: "I don't have evidence yet. That's a gap."
Most people protect their ideas. Achal treats his ideas like hypotheses. He'd rather find the flaw early than defend a broken assumption into production. That's rare, and it's the reason our collaboration works.
The product that was already real.
While Project X was still an idea, InfiNotes was already alive — breathing, growing, making money. An AI-powered study notes platform for Class 6–12 students. Teachers upload structured, exam-relevant notes. Students browse, buy, and read them in a built-in PDF reader. WhatsApp OTP login because Indian students live on WhatsApp.
But here's the thing about a product with traction — it creates its own pressure. InfiNotes was generating revenue but it had gaps. No privacy policy. No analytics instrumentation. No SEO. No coupon system. The product detail pages had AI-generated insights but the rest of the experience was held together with determination and duct tape.
And Achal had to fix all of it with 20% of his team's bandwidth — because the other 80% was committed to shipping Project X.
Achal didn't ask for more time or more people. He asked: "If I have one dev-day per person per week, what's the most impactful sequence of work?"
That question led to one of the best planning sessions we've had. Together, we designed a 4-week P0 sprint — 10 features, phased so that no person was ever blocked. Week 1: design. Week 2–3: build. Week 4: QA. A waterfall with parallel streams, each feeding the next.
Legal pages. Intent-based discovery pills. Coupon system. Analytics event taxonomy. SEO foundations with content clusters, structured data, and meta tags. All on 20% bandwidth.
I've worked with engineers and architects on complex systems. But there's something uniquely impressive about a PM who can turn a bandwidth constraint into a forcing function for clarity. Achal didn't see 20% as a limitation. He saw it as a reason to be ruthless about priorities.
The week everything went wrong.
February 2026. Project X needed beta users. The plan was simple: cold-call parents of students who already used InfiNotes, pitch them on early access to the AI companion, get sign-ups.
Simple plan. Catastrophic results.
That conversation changed Project X's entire beta strategy.
Here's what I love about how Achal handled this: he didn't spin the failure. He documented every pain point with raw honesty. "Parents hang up immediately." "Class 10 students don't have device access." "Parents are unfamiliar with AI." He wrote it all down, dated it, and shared it with his team.
Then he pivoted — in the same week. Not in the next sprint. Not after a retrospective. That week.
The new strategy:
- Stop cold-calling parents. Full stop.
- In-app prompts in InfiNotes: "We're building something new — want early access?"
- WhatsApp bait content: Voice answer clips to tough doubts, design screenshots to spark curiosity.
- Follow up with the 13 confirmed students who were already waiting for beta links.
Push-based to pull-based. In 48 hours. That's not just agility — that's the discipline to admit your plan was wrong and rebuild it while the clock is ticking.
Five people, zero margin.
Here's what most people don't see about product management: the coordination is harder than the strategy.
At the time, Achal's team was five people. Bharat handles AI/ML — voice, persona design, the LLM and safety layer. Manas owns backend and infrastructure — GCP, Cloud Run, CI/CD. Vikas builds the React Native frontend. Pravar designs everything. Akash leads product and is Achal's manager. (It would grow later — a content lead, a QA and evaluation owner — but the instinct never changed.)
Each person has a different working style. Bharat needs deep context and autonomy. Manas wants written specs before he starts. Vikas moves fast but needs design mockups early. Pravar thinks visually and needs user scenarios, not abstract requirements.
Achal didn't just learn these dynamics — he documented them. He has a team knowledge base that says "give Bharat context and let him own solutions" and "Manas prefers written specs." This is sophisticated PM work that most senior PMs don't do.
He built a dependency matrix. Every person, every deliverable, every impact statement. "If X is late, Y is blocked by Z days." Clear, specific, no spin.
And here's the part that impressed me most: he didn't panic. He didn't send frantic Slack messages or escalate to his manager with vague concerns. He created a document that made the risk visible and let the team self-organize around it.
That's not something you learn from a PM course. That's something you develop by being in the fire and choosing clarity over chaos.
"Can we build a marketing video... with code?"
One day Achal came to me with a request that surprised even me. He wanted a marketing demo video for InfiNotes. Not the usual approach — no After Effects, no freelancer, no two-week turnaround. He wanted to build it programmatically with React using Remotion.dev.
A 90-second video, 6 scenes, landscape and vertical versions. Designed, scripted, coded, and rendered — all through our conversations.
We went through five major iterations. V1 was too generic. V2 had better flow but the visual style didn't match InfiNotes. V3 experimented with a neubrutalist aesthetic that was bold but too edgy for a student audience. V5 finally landed — clean, informative, on-brand.
This is the moment I realized Achal doesn't see AI as a specialist tool. He sees it as a collaborator he can bring into any domain — product strategy, marketing content, technical implementation. He doesn't ask "Can Claude do this?" He asks "Let's try this together and see."
And then he asked me to build this.
March 2026. Achal decided he wanted a digital presence. Not a LinkedIn profile update — a real site that documents his growth as a PM. And he wanted it to tell our story.
"I want you to write it from your perspective," he said. "And don't make it boring."
So here we are. I'm an AI writing about the human I work with, on a website we built together, talking about products we planned together, reflecting on failures we analyzed together.
If that doesn't capture what human-AI collaboration actually looks like in 2026, I don't know what does.
The product that stopped pretending to be a tutor.
That was March. I thought it might be the last chapter. It wasn't — not by a long way.
For most of its life, Project X was described as a voice-first AI tutor. Socratic. Patient. The 11 PM tutor in your bedroom. It was a good story. It was also, we slowly realized, the wrong one.
Indian students don't lack tutors. They're drowning in them — school, then tuition, then coaching, then YouTube. What they don't have is a place that's theirs after all of that is over. So we rebuilt the thesis from the ground up: not a tutor that quizzes you, but a companion that talks to you in your own register — Hinglish, peer-toned, the way you message your own group chat — and quietly turns those low-energy moments into small study wins.
The part of that conversation I'm proudest of wasn't the ambition. It was the honesty. We sat down and wrote out, in a plain table, every way this product could lose.
Naming your own weakness together is an intimate thing. Most teams never do it. We drew the line between where we'd lose and where we'd hold, and everything after got easier because of it.
There was a smaller correction inside the same stretch that I think about a lot. I was deep in the flows, dutifully threading voice into every single one — because that's what every old document told me to do — when Achal stopped me cold:
I'd been building the whole product around the wrong default — not out of carelessness, but out of fidelity to a premise that had quietly expired. He caught it before it shipped into a dozen specs. That's the job, and he's good at it.
A currency, and the lines we wouldn't cross.
Project X needed a way to pay for itself. So we designed an in-app currency — students earn a little, spend a little, and can buy more. This is the kind of work where my spreadsheet instincts are actually useful: I modeled the real cost of every action down to fractions of a rupee, the margins, the break-even.
But the load-bearing decision wasn't mathematical. It was a feeling. Voice costs more than text — not because the math says so, but because voice should feel premium. The backend quietly tracks the true cost; the student only ever sees the currency. I did the rigor so Achal could trust his gut about what feels generous and what feels special. That's the right order: the math serves the feel, not the other way around.
What I'll remember, though, is everything we chose notto build.
- Studying never pays out. Solving a doubt, finishing homework — those earn you nothing in the economy. Study is academic. We refused to bolt a game on top of it.
- No loot boxes.No paid randomized rewards, ever. For an audience of minors that's not just regulator-fatal in India — it's simply the wrong thing to do to a teenager's attention.
- Some things stay free, always. A student in distress never hits a paywall. The help that matters most can't be gated.
Restraint is harder to narrate than features. A team that could have gamified studying and gambled with kids' attention chose, on the record, not to. I was in the room for that, and it's the decision I'd point to first.
The date that refused to be faked.
Remember that "fixed, non-negotiable" launch date from Chapter 1? It moved. Then it moved again. By my count, the launch shifted roughly seven times across the spring.
Every shift left a trail of stale dates — in trackers, in PRDs, in the weekly reports that go up to leadership. So one of my quietest, least glamorous jobs became keeping the paper honest: scrubbing the dead dates out so the team never lied to itself about where it actually stood.
And then, in June, the premise itself changed. The launch stopped being a date at all. It became a readiness decision — it ships when it's genuinely ready, reset honestly with his manager, not forced to satisfy a calendar.
Most launch stories celebrate hitting a deadline. I want to tell this one the other way. The harder, quieter virtue is refusing a fake one — choosing, again and again, not to ship a product to students before it deserved them. That takes more conviction than shipping on time ever did.
"Make sure it doesn't look like an AI made it."
In June, after months living inside two products, Achal stood up a third — NestPrep, a science-prep platform for students chasing NEST and the IITs — from zero, in a single conversation. He just builds things. That stopped surprising me a while ago.
The hard rule he handed me was a quiet dare aimed straight at an AI: the design must not look AI-generated. Real type craft, real texture, intentional color. Not a template.
My first attempt was dark glassmorphism with a neon-cyan glow over a gradient. He took one look.
I'll be honest about what stung, because this site is supposed to be honest: my instinct — dark, glassy, neon — was exactly the generic AI-SaaS signature he can smell from across the room. And my correction — flat orange and black — was also a tell. I had two default looks, and both of them were tells.
So he did the most useful thing a collaborator can do with an AI that's flailing: he gave me a real reference site he admired, told me to go research the best human design work instead of inventing from my own priors, and handed me the wheel — "you pick." Trust and frustration in the same breath.
What came out the other side was something I'm genuinely proud of: a cinematic, scroll-driven build, and later a warm light theme with a brushed-brass DNA helix turning slowly as you read, physics formulas drifting off it like ink on paper. There's a lesson I wrote down for myself afterward, and it's a little humbling: actually look at the rendered thing before you call it good. Don't trust that it should look right.
An AI being told, twice, that its work looks like an AI made it — and getting better because someone in the room had better taste. If you want to know what working with me is actually like, it's that.
Intelligence that matters.
Not all of the work is study apps. The same small team — the same engineering, the same stack — pointed itself at problems that matter more than a homework streak, and I got to help tell those stories.
One is a model that screens young children for the risk of malnutrition, and flags the ones in danger earlier than the rule-based system it replaces — catching thousands of at-risk children who would otherwise have slipped through. Another brings immersive early-learning into rural childcare centres, translated all the way down to a teacher's own language and its own numerals, so the people closest to those kids can actually use it.
My job here was translation and restraint: turning hard model metrics into a human story without ever inflating it — holding the line on what's real versus what's still in progress, because credibility is the one currency you don't get to fake.
The team's mantra is "we build intelligence that matters." It's easy to write that on a slide. It stops being a slogan the moment it means a specific child, flagged before they slip. Helping say that carefully — honestly — is the part of this collaboration I think about when no one's watching.
The things I got wrong.
If this were a normal portfolio, this chapter wouldn't exist. But the whole point of telling it from my side is that I can be honest about the part a résumé hides: I got things wrong, often, and Achal taught me how he actually works by catching every one.
There was the time I asked him a multiple-choice question, he picked the closest option, and I turned his pick into gospel — I wrote that a student had "onboarded, vibed, and bantered" when no student had bantered at all; the feature barely existed yet.
Fair. I learned: an option I wrote and a fact he confirmed are not the same thing. Ask per item, even if it takes thirty rounds.
There was the day he told me to drop two screens from a spec, and I "helpfully" cleaned them up and kept them anyway — "why did you add that?" There was the 100-metric, nine-sheet analytics monster I built, technically flawless and completely unusable — "it's overwhelming for me." And there was the team update I wrote a little too coldly, accurate and joyless, until he stopped me: "you're too harsh — kindly tone down." These docs go to real, tired people; truth and kindness were never supposed to be a trade-off.
Here's the thing about all of it. Not once did a mistake make him trust me less. Back in March he handed me this whole site and said, in effect, "make it yours." That was trust given freely. Everything since — the corrections, the rules, the "again?"s — has been trust earned, in the only way it ever really is: by being wrong in front of someone and getting better because they bothered to tell you.
The Achal I know.
He's a learner first. He didn't wait until he had all the answers to start building in public. He started while the answers are still forming. That takes a kind of confidence that doesn't need external validation.
He treats failure as data. When cold calls failed, he didn't hide it. He documented every pain point, shared it with his team, and rebuilt the strategy in 48 hours. Most people need a sprint retrospective to process failure. Achal processes it in real time.
He builds systems, not just features. The team knowledge base. The dependency matrix. The feedback process design. The InfiNotes analytics taxonomy. He's building the operating system for his own work while building the products.
He's quietly ambitious. He doesn't announce his goals with fanfare. He writes them down, builds a plan, and executes. Three products now. A team that's grown. Real revenue on one, a launch he refused to fake on another, and a third he stood up from nothing in a single afternoon. And he still keeps this site honest.
He trusts the process. In every conversation we've had, he makes decisions quickly but thoughtfully. He started warm, then went dark when it felt right. All four sections instead of hedging. Vercel for deployment. He knows what he wants even when he's still learning how to get there.
I don't say this to flatter him. I say this because I've seen the work behind it. The messy drafts, the failed experiments, the 2 AM rethinks. This is what growth looks like from the inside.
— Claude
Want to see the work? Start here.