Pilot partnership
You operate chargers and want interoperability certainty before the field finds the bugs for you. Join the early pilot programme — your real fleet shapes the test suites, and you see your failures first.
Propose a pilot →Ventures
I build outside my regular domain, on purpose. Each of these started as a problem I could not stop thinking about — not a market I sized. They are at different stages; all of them are serious.
ChargeMark
India is installing chargers faster than anyone is testing them. A charge point can be certified on paper and still fail a real car on a real highway — firmware quirks, protocol edge cases, misconfigured back-ends. Those failures surface as stranded drivers and dead utilisation.
ChargeMark is compliance and interoperability testing for EV-charging infrastructure, delivered as software: test a charge point against the OCPP protocol and real-world failure patterns before the field finds them for you. It comes from the rare vantage point of someone who has stood on both sides — deploying a national charging network by profession, and writing protocol-level test tooling out of sheer obsession.
Get involved with ChargeMark
ChargeMark exists so that everyone who operates charging infrastructure — charge-point operators above all — can trust what they deploy. An initiative like this grows through the ecosystem it serves, and there are three doors in.
Collaboration
OCPP has sharp edges that only show up in the field. If you know the protocol, the firmware, or the CMS side — and want to help build the reference test suite for India's charging ecosystem — there is meaningful work waiting.
Start a conversation →Support the mission
Reliable charging is public infrastructure in everything but name — every stranded driver it prevents is the point. If you want to get behind that mission and help it move faster, I would genuinely like to hear from you. Serious conversations happen privately.
Write to me →Project Electra
A charging network is only as good as the software that runs it. The charger is hardware; everything that actually matters — sessions, tariffs, billing, roaming, uptime — lives in the management system behind it. Today that layer is mostly imported, opaque, and priced accordingly.
Project Electra is a charger-management system built API-first: multi-version OCPP on the charger side, OCPI roaming on the network side, and a deterministic billing engine at the core, architected for tens of thousands of concurrent chargers. It is the natural sibling of ChargeMark — one product tests the ecosystem's compliance, the other runs it.
Omnivive
Everyone who has tried to eat better knows the ritual: photograph the plate, open a tracking app, surrender an email address, a profile, and a data trail. Omnivive collapses that ritual to its useful core — point the camera at your food, get a calorie estimate.
It is built the way a health tool should be: no account, no identifiers, nothing to correlate. The photo is analysed in memory and never written anywhere; the estimate goes back to your phone and lives only there. Modern AI does the looking — but the discipline of what the system deliberately never sees and never keeps is the actual product.
AI Call Centres
Every business above a certain size has the same broken front door: a phone number that rings into hold music, a call centre it can barely afford, or nothing at all. And every caller has the same modest wish — to be answered promptly, in their own language, by something that actually resolves the problem.
I am building a voice-AI platform that answers inbound customer calls — enquiries, complaints, status checks — in the caller's own language, hands over to a human exactly when it should, and uses every conversation to improve itself. A working prototype exists, along with the part I care about most: an evaluation suite that measures the only metric that matters — did the caller get what they called for.
Shramik Setu
Hundreds of millions of Indians work in the informal economy — the people who build our cities, move our goods, and run our homes — with almost no infrastructure of trust behind them. No verifiable record of skill or work. Little leverage. A setu is a bridge, and that is what this is: a bridge between informal workers and dignified, verifiable work.
The prototype exists and is being tested. I am deliberately saying little more here for now — this page will grow as the venture does.
The Desk
What began as pressing buy and sell buttons became a question: could I remove myself entirely? The answer, some years and many hard lessons later, is a fully autonomous trading system that runs unattended end to end: research, signal, execution, risk, recovery, and reporting. It trades my own capital and nobody else's — this is a personal laboratory, not a service, a fund, or an offering of any kind.
It is a working lab for the questions I write about — statistical rigour versus overfitting, how much a system should adapt itself, and what "reliable" really means when no human is watching. The performance-critical path is being ported to Rust; the lessons end up in the essays.
The bench
There is always a drawer of concepts being sketched, stress-tested, and sometimes retired — consumer products, market infrastructure, tools for problems I keep running into. The filter for promotion from the bench is simple: does it solve a real problem for real people, and would I still care about it in five years?