This publication describes itself in one line: a publication and podcast about the people who build without an audience. The podcast is called Built in Silence, and its episodes, the homepage promises, ship when they’re ready. The position is deliberate. Most of the operators documented here spent years doing the work before anybody wrote about it, and the work was usually better for the quiet.

So it is only fair to open this filing with the sentence that argues against all of it, written by today’s operator in an essay titled, with very little room for negotiation, “Distribution Is the Whole Game.” “Building in silence until it’s ‘ready’ isn’t humble,” he writes. “It just means you’ve chosen to start your distribution from zero on the single worst day to do it.” The uncomfortable part is that the arithmetic is on his side. Most founders eventually meet the same morning: product finished, launch post written, room empty. The work that fills a room takes years, and nobody can start it on the day it is needed. His fix is a schedule, not a personality: “You build it for two years before you need it, so that on the day you finally have something to sell, there’s already a room full of people listening.”

This series documents operators who encode hard-won judgment into systems, most recently in our last transmission, an agency leader who fitted the missing fourth step, active recovery, to the oldest loop in management. Today’s filing, the eighteenth, concerns an operator only a few years out of an engineering degree, and a system of an unusual kind. The judgment he encoded is not about a market or a craft. It is about attention: what earns it, how it compounds, and how long it takes to arrive. He built the room first. Then he co-founded a company worth bringing into it.

The operator in question

Manav (most of the internet knows him as Tensor Boy, but first names are policy here, and he can introduce himself properly) came up through the part of Indian engineering education that runs on a single instruction. In the second year of his computer science degree at GL Bajaj, the whole batch was grinding data-structures problems for placement tests: “500 LeetCode problems, clear the OA, land the service company job, exhale.” He opened Kaggle instead and, by his own account, “didn’t look up for eight months.” His reasoning, from an essay called “Why I Chose ML Over DSA,” is the cleanest statement of his method. The placement grind is a filter, built so a recruiter can reject most of a pile before lunch, and it teaches almost nothing about building. “Skip the filter. Learn the work.” The Kaggle years ended with him a three-time Kaggle Expert.

The work followed at a pace that reads, on his profile, like a compressed decade. Machine learning engineer at Perception.ai. Large language model engineer at YogLeads, building open-source retrieval pipelines his profile credits with cutting costs by ninety per cent. Generative AI architect at Aary OM, advising startups on saving money with open-source models. Seven months at Innovaccer as an AI/ML engineer, on a clinical decision-support system and the machinery for serving models at scale, which is worth remembering for later. Diffusion-model research at 100xEngineers. A bootstrapped lab of his own, Pluto Labs, out of buildspace’s fifth season. Six-plus companies shipped into, by his count; seventy-five-plus hackathons as participant, mentor, judge and host; two TEDx stages.

And here is the tell, visible before any of the AI arrived: the audience came first even then. From 2020 he was a speedcuber, coaching children one-to-one on the Rubik’s cube in hour-long online lessons, and his profile lists two years as a cubing influencer who “promoted art online,” with a “strong Instagram following, engaged community.” Then community-manager roles, moderating Discord servers and running events, and a stint as a content editor that, by his own note, sharpened his writing. Before he had anything technical to say, he had learned how to hold a room around a cube. When he did have something technical to say, he already knew how to say it daily.

The Tensor Boy channel started in November 2023. By his own figures it now runs to 257,000-plus followers across platforms and more than ten million monthly views, with a newsletter, Tensor Protocol, read by ten thousand-plus developers and builders, and a list of brand partners (Apple, Warp, Notion and Y Combinator among them) that most media companies would frame. The method behind the numbers is deliberately unglamorous. He showed up every day and “documented instead of bragging”: “Here’s what I built, here’s what broke, here’s what I learned.” “Boring. Daily. It worked.” His LinkedIn summary reduces the editorial policy to four words: “no hype without receipts.”

He has also written, with unusual candour, about what the daily machine costs the person feeding it: “Build something, burn out, disappear, hate yourself for disappearing, force a comeback, burn out harder. Repeat.” Readers of our last transmission will recognise the shape. It is the three-step loop with no fourth step, run in public, and the essay it comes from is where his newsletter’s house rule was set: write only when there is something worth the reader’s time.

The room and the clinic

In December 2025 he co-founded Cevi with Theo, who runs it as chief executive; Manav is the CTO. The company builds AI agents for medical practices in the United States: agents that, in the site’s words, “answer calls, verify insurance, schedule visits, manage refills, and collect payments,” living inside the practice’s electronic health record rather than beside it, with 148-plus native integrations. The workflow library lists forty pre-built agents, from an AI receptionist to waitlist backfill for cancelled slots. Anything not on the list, the site offers to build as a custom agent, deployed “typically in under a week.” By Manav’s account, the company was profitable from month one.

The design principle sits on the contact page, in his words: “We don’t bolt AI onto broken workflows. We rebuild the workflow around what AI does best.” Theo supplies the diagnosis it answers: “brilliant clinicians drowning in admin.” Anyone who has watched an American practice run will know the layer he means: the phone queue, the eligibility check, the refill request, the prior authorisation, the bill nobody understands. A behavioural health group in Texas, quoted on the homepage, describes the result in operating terms: “They now handle the majority of our patient operations end to end, including calls, intake, insurance verification, and scheduling, with minimal human involvement.”

Two inheritances show in the product. The first is technical. The Innovaccer months on clinical decision support and model serving, and the YogLeads habit of making retrieval cheap with open-source parts, are the CTO’s résumé written into a company that has to be EHR-native, reliable and affordable for a small practice. The second is the room, and here an honest reader will object. The quarter of a million are mostly engineers, students and builders; Cevi sells to medical practices. Nobody in the Tensor Boy audience is buying refill automation.

But a company draws more from its distribution than customers. It draws engineers who already know how its CTO thinks, partners who have watched him ship in public, invitations to rooms a sales team could not buy its way into, and the benefit of the doubt on a first call. Cevi’s own contact page routes hackathon sponsorship and public speaking to its CTO. The room is not the sales channel. It is the reputation the sales channel runs on. And he keeps giving the room something to do: Tensor Hack’s short email, The Drop, sends out the best new AI hackathons, jobs and opportunities, each one “checked by a human first.”

Notes for anyone building the room

Strip away the follower counts and the transferable rules read like this:

Start the room before the product. Distribution compounds on a clock nobody can speed up. Launch day is not the start of distribution. It is the first day you need it, and the worst day to begin.

Document, don’t perform. “Here’s what broke” ages better than any victory lap. Receipts are a format, and an audience built on them is still there after a bad quarter.

Clear the filter, but never confuse it with the work. Credentials that exist mainly to reject people are worth passing and never worth worshipping. Depth in the thing that is about to matter compounds; a ranking does not.

Rebuild the workflow before you automate it. AI bolted onto a broken process automates the breakage. The agent should fit the work as it ought to be done, inside the systems where it is actually done.

Ask. His essay on the subject ends in four words: “Dignity is overrated. Ask.” The cold message costs a minute. The opportunity that quietly expires because nobody sent it costs far more.

The quiet objection

So, back to our own masthead. Built in Silence was never meant as an instruction to hide. It describes the years of unwatched work that happen before anyone writes about them. Read closely, Manav’s career contains exactly those years: eight months of Kaggle notebooks nobody asked to see, at a point when his whole batch was walking the other way. The difference is that he published the notes as he went. On the evidence of this filing, the two positions are closer than either slogan admits. Do the quiet work, and let people watch.

So: the agents are at cevi.ai, the essays and the newsletter at tensorboy.com, and the room is open to anyone who wants practical signal on AI. His own summary of the whole enterprise is lower-case and exact: “but the real goal is not reach. it’s trust.” Most founders build the product and then go looking for a room. He built the room, then built something worth its trust. For a publication about people who build without an audience, that is an awkward thing to concede. We are putting it in writing anyway.

— END TRANSMISSION 06X

Read the next one.

We’re publishing often at the moment. One email when the next one lands, if you ask for it.

Ask for the one email →