About
Hi — I'm Ishaan.
I've spent the last seven years building machine-learning systems that run for real: anomaly and fraud detection at a cybersecurity company in Mumbai, then — after an MSc in AI brought me to London — pricing and forecasting models at a debt-purchase group, where I now build the LLM side of things: self-hosted models, document pipelines, and agentic systems, all inside a regulated industry.
Outside work I build my own. Alter — an AI second brain that turns the scatter of things I capture into one private, searchable wiki. Spiral — a design for AI filmmaking that works the way filmmakers actually do, in revision loops. And the two short films that came out of it, thisness and Samudra Manthan.
What I want next is simple: to work on agents and multi-agent systems full-time, with a team that treats social coordination as a design problem — the rules and structures that could actually fix it. The version I keep sketching is a game, simple and fun, where everyone does what they already wanted to do and the wants quietly add up to a society that runs — one that rewards people for being themselves. Someone has food to spare. Someone fancies a quick walk. Someone, a few streets over, is hungry. Those three wants already fit together; what's missing is the system that notices, matches them, and gets the food into the right hands safely — everyone a little happier for having done exactly what they felt like doing.
That's the whole of it, really — build multi-agent frameworks, design games, make films that make me feel something, and be somewhere warm with a tiramisu and a limoncello spritz while the system runs without me.
The longer answer
Why agents — and why films.
Everything I'm drawn to has the same shape: many small parts, nobody in charge, and — when it goes well — one coherent thing on the other side. Nobody sets the price of milk, yet every morning it has one. No ant plans the colony. Trillions of cells, no manager, and here you are, reading.
For most of history you could only admire that pattern. The reason agents have my full attention is that you can now engineer it: write the rules, set the incentives and penalties, and watch what behaviour actually emerges. When the emergence is wrong, you don't scold the agents — you redesign the game. And I mean "game" precisely, not poetically: a good game is a set of rules under which the behaviour you want appears on its own. Multi-agent engineering is exactly that job. It's what I'm experimenting with, in the open, here.
Alter is the same idea aimed at knowledge — a scatter of notes, captures and half-thoughts becoming one wiki that stays coherent without me curating it. The films are it aimed at feeling — thousands of generated frames that have to add up to one story.
Vipassana keeps me honest about the limits. Sit still long enough and the "one thing" you call yourself turns out to be a crowd of parts, briefly agreeing. So I hold two instincts that don't resolve: the engineer who wants to author the rules, and the practice that says let the authorship go. I've stopped trying to settle it. The friction is where the interesting work seems to come from.
That's the name, too. null — the void, the unformed, the many before they mean anything. sutra — the thread that binds. This whole site is the move the name describes: suturing meaning into the void.
Right now
Open to Senior AI Engineer / Forward Deployed Engineer roles.
Especially with teams building agentic systems or generative media. If something here resonated — about a role, or just to compare notes — write to me.