Assurd Techlabs, Pune
For the past seven years, Sourabh's work has centered on architecting and scaling machine learning systems across the United States and India. As a Senior ML Architect, his focus was on the deep infrastructure layer — optimizing distributed training, cutting latency, and ensuring models held up under massive compute demands.
Recognizing that high-performance software is ultimately at the mercy of the physical hardware it runs on, he identified a critical industry vulnerability: the deployment of incredibly complex AI models on top of hardware ecosystems that operate almost entirely on blind trust.
To solve for this physical layer, Sourabh founded Assurd Techlabs. Today, he is replacing trust with cryptographic and physical proof; engineering forensic test rigs and thermal validation systems to certify high-end compute hardware before it ever hits a production environment. Alongside building Assurd Techlabs, Sourabh takes on selected fractional and part-time ML infrastructure engagements, helping teams deploy production ML systems and improve cloud infrastructure efficiency.
Most hardware transactions run on trust.Trust breaks. Systems don't.
We are building the definitive verification layer for high-end compute. By combining forensic physical inspection with immutable digital records, we ensure that every piece of hardware entering your infrastructure is mathematically and physically proven to perform.
Deep-level validation of compute components to ensure pure authenticity.
Pushing limits on open-air chassis test rigs to uncover any physical degradation.
Immutable cryptographic records generated for every single hardware unit.
Every unit is physically triaged and stress-tested on our custom, open-air hardware rigs. We push silicon to its absolute thermal and computational limits to expose hidden degradation, micro-fractures, and voltage instabilities. This rigorous physical validation leaves zero room for guesswork, ensuring enterprise-grade reliability.
Thermal and power telemetry reveals the undeniable truth under extreme load. By capturing high-frequency data across voltage regulators and compute dies, we establish a precise baseline of hardware health. This telemetry is cryptographically signed and bound to the unit's unique physical identity.
Macro inspection validates component authenticity down to the silicon level. We utilize high-resolution optical analysis to detect counterfeit PCBs, altered serial numbers, or unauthorized micro-soldering. The final output is a public, undeniable certificate of authenticity, not just a vendor's promise.
Every hardware unit certified by Assurd generates an immutable, publicly verifiable record. This isn't a sticker—it's a mathematical guarantee.
Infrastructure · ML Systems · Cloud Architecture
Assurd is a long-horizon infrastructure play. While the certification platform scales independently, Sourabh takes on a small number of consulting engagements — typically two to three concurrent part-time contracts — with companies solving hard ML infrastructure and cloud architecture problems.
Currently accepting part-time remote engagements (20 hrs/week). Focus on ML infrastructure, cloud migration, cost optimization, and retrieval systems.
The Problem: Total RAID failure caused complete loss of database connectivity across legacy .NET/C# application stack with no disaster recovery in place.
University of North Carolina at Charlotte
Specialized in scalable machine learning architectures and distributed computing. First in the family to leave the country and earn a Master's degree.
University of Pune
First-generation engineer. Built the deep algorithmic and systems engineering foundation that drives Assurd's infrastructure today.