Fractional CTO & Systems Architect
Twenty years going from precision to precision at scale.
The first half of my career was silicon — automotive ECUs built to MISRA and ISO 7637, engine control for a defence vehicle prototype, Class 2B medical devices, and an open-source build system Carnegie Mellon adopted for their Iris lunar rover's flight control build.
The second half was scale — fulfilment for one of India's largest e-pharmacies, 10,000+ warehouse robots running 24×7 across three continents, B2B payment rails settling T+0 with statutory deductions at source.
Companies bring me in when technology has become the constraint: the platform won't scale, the burn won't come down, delivery keeps slipping, or diligence has flagged the architecture. I've built two companies and led engineering in four more, so I've usually met the failure mode before — and I work at whatever depth it takes, from the board conversation to the critical-path bug.
| −76% | infrastructure TCO, with throughput and operational stability up |
| −50% | cost of engineering delivery, with output held constant — AI-assisted workflows, automated MLOps, and tooling that removes the work rather than the people |
| 30× | fewer customer on-call issues — roughly 6,000 a year to roughly 200 — removing about $3M in annual cost |
| 50 → 80% | delivery predictability across a 75-engineer organisation |
| 100× | navigation accuracy, 1000 ppm to 10 ppm, across a fleet of 10,000+ warehouse robots |
| $1.5M | monthly transaction value settled T+0, split at source, GST/TDS/TCS automated |
| 17 | countries still running medical devices I designed, a decade on |
A warehouse robot needed a real-time kernel on a TI AM335x. AWS FreeRTOS had no working port for that chip with the TI interrupt controller, so I wrote one, along with the build system around it, and put both in the open. In 2021–22 Carnegie Mellon adopted the build system for the flight control build of their Iris lunar rover, which flew on Astrobotic's Peregrine Mission One in January 2024. Both are in GitHub's Arctic Code Vault. — c_source_tools
A hardware company's device authentication had weak entropy and had been broken. Cloned transmitters were pairing with genuine receivers, and every clone was revenue that never arrived. I replaced it: encryption-as-a-service in the cloud, custom cryptographic JWTs, keys generated on dedicated hardware at the point of manufacture, and a three-way handshake binding server, mobile app and firmware. The fix reached back into the factory, which meant coordinating five teams that don't normally share a schedule.
Counterfeit consumables were entering the supply chain for a Class 2B wound-healing device I'd designed. Patients were being treated with spurious parts on equipment that had to work, and the complaints came back as the device is broken. We closed the loop end to end — connected device fleet, mobile app, cloud backend, and authentication built into the manufacturing line — then layered ML across the fleet for early warning on field failures. Those devices are still running in top hospitals across India and exported to around 17 countries.
Most engineering teams calibrate their rigour by habit rather than by consequence.
I've built systems where being wrong costs a limb, and systems where being wrong costs a rounding error. The teams that struggle are the ones applying medical-device process to a checkout page, or checkout-page process to something that touches a human. Knowing which one you're building is most of the job.
Currently Fractional CTO to a deep-tech company applying computer vision and spectroscopy to agricultural quality assessment at national scale, and open to a small number of additional engagements.
If technology is what stands between a company and its next stage — whether you're the founder, the investor or the board — get in touch.