Building the experimentation layer for AI agents - because every team is still doing it by hand.
Mayin Joshi runs SquareDiff, an autonomous agent optimization platform that lets AI teams stop fiddling with prompts, models, and tools by hand. The platform experiments for you - finding the fastest, cheapest, most reliable configuration of any given agent - while you actually build the product that matters. He co-founded it while simultaneously serving as Growth Marketing Lead at Productboard, where he was the youngest hire in company history when he joined in September 2025.
By the time Joshi moved to San Francisco at 18, he had already run four companies. Not side projects - real operations with real revenue. A sneaker marketplace that scaled to $680,000 in gross merchandise value in three months. An agency that billed five figures monthly at age 16. A CRO and analytics firm whose client list read like a streetwear and consumer tech who's-who: Crypto.com, Culture Kings, Abercrombie, Lulus, WeMod.
The thread connecting all of it: a compulsive need to understand how systems work and then make them work better. He taught himself bot automation and proxy management as a pre-teen to win sneaker drops. He studied retrieval-augmented generation and agentic architecture in his teens. He moved to SF with angel commitments in hand. Now he's building infrastructure for the AI era.
Context is key.
- Mayin JoshiThat three-word GitHub bio is not accidental. It doubles as a mission statement. His technical obsession with RAG systems - the retrieval mechanisms that give AI agents memory and situational awareness - informs everything from SquareDiff's architecture to the way he thinks about growth marketing. What does the system know? What context is it missing? How do you close that gap?
At Productboard, Joshi works cross-functionally across product, data, engineering, and GTM to design and execute new growth initiatives. He is not running campaigns. He is building the machinery that makes growth reproducible - a distinction that matters to him considerably.
It started with sneakers. Most stories about teenage entrepreneurs involve lemonade stands or lawn mowing. Joshi's involves bot automation software, proxy networks, and secondary market logistics. At twelve, he figured out that limited-edition sneakers were a denominated asset - predictable scarcity, liquid market, quantifiable spread between retail and resale. He put in $180 and started learning.
By thirteen, he had six-figure revenue. By fourteen, six-figure profit. He had also taught himself the technical infrastructure behind sneaker copping: how automated purchasing bots work, how proxies obscure requests, how to manage inventory at scale. These were not trivial skills. The sneaker resale market is adversarial - platforms actively fight automation, and only operationally sharp players survive.
When he was sixteen he took that operational knowledge and formalized it. Kick Vault launched as an invite-only B2B marketplace connecting secondary-market retail stores with bulk streetwear sellers. It was not a consumer app - it was infrastructure for the industry he already knew. Three months in: $680,000 in GMV. The business had legs, and Joshi had proven he could build something other people needed.
Around the same time, he launched an organic social marketing agency for consumer tech and gaming companies - clients like Crypto.com and WeMod - and scaled it to five-figure monthly recurring revenue at sixteen. Then came Expocord, a conversion rate optimization and data analytics agency, which added Lulus, Abercrombie, and Culture Kings to the roster.
He graduated high school early - at seventeen - which tells you something about how he relates to conventional pacing. At eighteen, he moved to San Francisco. Not for a job. Not for college. To be where the things he cared about were happening.
He attempted a fintech startup called Lop, raised angel commitments, then hit legal barriers and paused. That story - raise money, hit a wall, learn, pivot - is not a failure; it is what building things actually looks like. Most people who haven't done it think the hard part is the idea. The hard part is navigating the wall you didn't see coming.
He spent the next period consulting for seed-to-Series-C startups across robotics, consumer AI, and B2B SaaS, and simultaneously going deep on generative AI architecture. Not reading blog posts - studying RAG systems, agentic retrieval, multimodal models. Building things to understand how they work. By 2025 he had founded both SquareDiff and Rylo (a fashion shopping app using multimodal AI and personalization), and joined Productboard as their first growth hire.
The through-line from sneaker bots to autonomous agent optimization is not as strange as it sounds. Both are about making systems work reliably at scale in adversarial or high-variance environments. The domain changed; the approach did not.
Every AI team building agents faces the same grind: which model? Which prompts? Which tools? In what combination? The answer changes as models improve, as costs shift, as latency requirements tighten. Right now, most teams run these experiments manually - a slow, expensive, non-reproducible process that distracts from actually building the product.
SquareDiff automates that loop. Feed it an agent, and it systematically experiments across prompts, model choices, tool configurations, and more - finding the version that's fastest, cheapest, and most reliable for your specific use case. The platform runs the grunt work. The team runs the product.
It sits in the AI infrastructure and developer tools space, targeting AI teams who are past the prototype stage and into the optimization problem. That is an increasingly large and increasingly frustrated audience.
Joshi's co-founder at SquareDiff is Mayank Singamreddy. The technical foundation reflects Joshi's own period studying agentic retrieval - he built the thing he needed to understand first.
Productboard is a $257.7M-funded product management platform - Series D, based in San Francisco - built around the idea that product teams need a single place to understand customer needs, prioritize features, and align the company around a roadmap. It competes on clarity and alignment in a space where most product decisions still happen in Slack threads and shared spreadsheets.
When Joshi joined in September 2025 as Growth Marketing Lead, he became hire #1 in the growth function - meaning he was building the function, not filling a seat in one. His mandate: design and execute growth initiatives full-funnel, working with product, data, engineering, and GTM. The kind of cross-functional generalist role that either suits someone perfectly or destroys them. Joshi, predictably, is suited for it.
His take on inbound GTM in 2025 is characteristic: "it's about signals, demand, and systems - not just content." Most growth operators are still wiring together content calendars. He is thinking about what signals indicate purchase intent, how demand is structured, and what systematic process captures it reliably. Different frame, different results.
The fact that he runs this role while simultaneously co-founding SquareDiff is the detail that most people find hard to hold. It is, at minimum, evidence of unusually high energy and a very efficient relationship with time.
Context is key.
GitHub bio - and something close to a life philosophyInbound GTM in 2025 is about signals, demand, and systems - not just content.
On growth marketing, shared via LinkedInBuilding AI products isn't about chasing hype - it's about solving real customer problems.
On AI product development