I’m founder, b.

Himanshu

The hard part is rarely the model.

I'm Himanshu. I'm 21, and I run Marl Studio, a game studio building its own game engine.
Before that I was CTO at AuraX, built agents at Elip AI (YC W25), and grew SynthAILabs to 11,000+ students.

Each time, the model was the easy part. What took the work was everything around it: memory, routing, money, and getting the thing to people.

My latest are ashlar, a Rust app framework, and AOTT, our first game.

Away from code I read books and draw.

What I built, and what it taught me

Marl Studio

Founder, Aug 2026 to now

The problem. Game engines are heavy to run and closed to AI agents. A small studio that wants to ship games quickly needs an engine that is fast, simple to run, and one an agent can drive.

What I'm building. Marl Studio is a game studio building its own engine. Our first game is AOTT: hold the last wall, a free action game in the browser, played on the keyboard with nothing to install. The first version of the engine is public as hull-dep; I'm redesigning it, and the new one is private because it carries proprietary tech.

What it's teaching me. Distribution works best when it is built into the product. A free game that runs in a browser tab, with nothing to install, is its own demo.

Elip AI (formerly Tejas AI, YC W25)

Software and AI engineer, Mar to Aug 2026

The problem. Job hunting is mostly logistics, and Elip had to handle it inside WhatsApp: one text box, no settings, no forms, and a tired person on the other end.

What I built. A FastAPI agent with a router and specialist sub-agents for resumes, reviews and job matching, with memory across Redis, Postgres and S3. Under it, hybrid search on OpenSearch over Go services indexing jobs from about 60 ATS platforms and job boards.

What it taught me. None of the hard parts were model problems. They were memory, deterministic routing, and knowing when a requirement is a filter rather than a similarity score.

A router that occasionally decides to freestyle is a router that occasionally loses someone's resume.
From my Elip AI write-up

AuraX

AI researcher and CTO, Dec 2024 to Dec 2025

The problem. Fashion e-commerce needs on-model photos of every garment, in every size, on every body type. A photoshoot per SKU does not scale.

What I built. An aesthetic fine-tuning loop for fashion images: generate, score for aesthetics and garment accuracy, feed the scores back, repeat. In blind tests, people picked ours as real over Google's and OpenAI's image models. On Nautinati, Aditya Birla Fashion's kidswear brand, our images got a 3% click-through rate, against 1.5% for studio shoots. The try-on research is written up as Flux-VTON+.

How it ended. We raised ₹22 lakh from IIIT Hyderabad's CIE, IIM Bangalore's NSRCEL and MeitY's TIDE 2.0, and closed deals with Mensa Brands and Aditya Birla Fashion. When the next round didn't come in time and the big labs moved in, we turned down ₹28.5 lakh more. It would have kept us alive, not ahead. The full story.

SynthAILabs

Founder, Dec 2023 to Feb 2025

The problem. Engineering students with the ability but not the ramp take far too long to become employable.

What I built. Free AI-made learning material built around projects, which grew into a community of 11,000+ students. Alongside it I co-founded OpenCode Developers, a club that took 500+ students through competitive programming in one summer instead of the usual six months.

How it ended. I wanted it to stay non-profit and couldn't fund it, so I wound it down.

The long version →

Freelance or advisory work

Marl Studio is my full-time work, and I take on a few freelance builds and advisory roles alongside it: Rust desktop apps, AI agents and image models, or go-to-market for a product that already works. Start with a 15-minute intro call, or write to hyattherate2005@gmail.com.

Book an intro call