Lab
What I'm currently building, reading, and learning. A living document — updated irregularly, honestly.
Building AI copilot demos
Developing browser-based AI engineering tools — starting with FEA optimisation for semiconductor wafer chuck design. Exploring how LLMs can close the loop on simulation iteration cycles. See demos
Exploring NVIDIA Omniverse
Going deeper into Omniverse for physics-based simulation workflows and Isaac Sim for robotics. Particularly interested in the bridge between USD-based scene composition and engineering simulation pipelines.
Reading on physics + LLMs
Consuming papers and articles on the intersection of physics-informed neural networks and large language models. The question: can LLMs develop physical intuition, or will they always need a solver in the loop?
Planning next trek
Looking at routes in the Western Ghats for early 2026. Coorg or Kudremukh — still deciding. Also keeping up with German lessons, sehr langsam.
TIL Log
Small, specific things worth remembering.
USD (Universal Scene Description) by Pixar is becoming the lingua franca for 3D engineering data — NVIDIA, Autodesk, and Siemens are all converging on it. It's essentially what PDF is to documents.
Physics-Informed Neural Networks (PINNs) encode governing PDEs directly into the loss function. They don't need labelled data — physics is the supervision signal. Neat inversion of normal ML thinking.
Star CCM+ has a Python scripting layer that lets you drive simulation geometry and mesh parameters from external code. This is the hook needed to build a proper closed-loop AI optimisation workflow on top of CFD.
The DiSC "Dominance" trait predicts comfort with risk and ambiguity, not just assertiveness. High-D people don't dislike process — they dislike process that slows outcomes. Reframe accordingly.
Vercel's Edge Network caches static HTML at 100+ PoPs globally — a purely static site on Vercel can serve sub-50ms TTFB for most users worldwide. No backend complexity needed for a fast personal site.
Kudremukh means "Horse face" in Kannada — the peak's silhouette from a certain angle. The name is 2,000 years old. Karnataka's forests are older than anything I'll ever build.
Reading list
Currently reading
The Coming Wave
How AI and synthetic biology will transform the next decade — and why containment is the defining challenge.
Currently reading
Chip War
The geopolitics of semiconductors. Directly relevant to the simulation and CAD world I've spent 20 years in.
Up next
Deep Learning for the Life Sciences
Curious whether the transfer-learning patterns in bio translate to physical simulation workflows.
Finished
The Lean Startup
Re-read this with fresh eyes after the Mercedes SAP S/4 greenfield. The validated learning loop is universal.
Finished
Turn the Ship Around!
Leader-leader vs leader-follower. Best leadership book I've read for people managing large technical teams.
Finished
Thinking, Fast and Slow
System 1 and System 2 thinking shapes how I frame every change management and adoption conversation now.
Side projects
Active · 2026
AI Copilot Demo Suite
A growing collection of browser-based AI engineering tools. FEA optimisation live now; materials, thermal and yield copilots in progress. Pure HTML/JS — no backend.
Research · 2026
LLM + FEA Closed-Loop Optimisation
Exploring whether an LLM can reliably propose FEA parameter adjustments — not as a solver, but as an intelligent search heuristic guided by simulation results. Inspired by the wafer chuck copilot.
Exploration · 2024
PLM Data to Digital Twin via Omniverse
Prototyping a pipeline from Siemens Teamcenter product structure to a live USD-based digital twin scene in NVIDIA Omniverse. Early stage, nothing to show yet.