AI Architect at Oracle's AI Centre of Excellence, based in Spain. I design, deploy, and benchmark GPU-accelerated AI/ML & HPC infrastructure across the full NVIDIA AI Enterprise stack — and build open-source governed-AI systems on the side.

About

Senior architect with 20+ years designing and deploying high-performance AI/ML, HPC, CAE, and GPU-accelerated cloud infrastructure across global industries — leading complex architecture engagements, cloud migrations, and GenAI platform enablement for research, enterprise, and hybrid environments.

My path runs from automotive CAE and financial-services HPC to enterprise GenAI on the cloud. Today I focus on reference architectures, performance baselines, and guardrails for GPU-accelerated GenAI/HPC — and on publishing open-source frameworks, benchmarks, and research along the way.

Expertise

Designing, deploying, and benchmarking large-scale AI/ML & HPC environments — hands-on across the full NVIDIA AI Enterprise software stack.

  • NVIDIA AI Enterprise — NIM, NeMo, Triton, TensorRT, TAO, RAPIDS / cuVS, DeepStream, cuOpt
  • GPU infrastructure — large-scale A100 / H100 / H200 clusters, multi-node GenAI & distributed LLM training
  • Benchmarking & optimization — throughput, latency, scaling, and cost/token across the inference & training stack
  • Cloud & platform — OCI / AWS / GCP / Azure, Kubernetes, Slurm, Terraform & Ansible IaC
CUDA · NCCLvLLM · TensorRT-LLMLustre · GPFS RDMA / RoCEPrometheus · GrafanaPython · Linux Automotive CAEFinancial HPC

Experience

AI Architect — Oracle
AI Centre of Excellence · reference architectures, benchmarking & guardrails for GPU GenAI/HPC
2021 — Now
Senior Professional, Emerging Technologies — DXC
HPC & emerging-tech consultant across financial services, aerospace & automotive
2018 — 2021
HPC Analyst — Citi
Financial Engineering Research Group · trading & risk modeling on simulation grids
2016 — 2018
Lead HPC Solutions Developer — Tata
CAE Research Group · automotive simulation clusters (LS-DYNA, Abaqus, Fluent, Nastran)
2008 — 2016
Senior Linux System Administrator — Sankalp
Enterprise Linux infrastructure, web/mail servers & team management
2007 — 2008
Programmer & Academic Mentor — Vindhaya
Taught computer science, lab sessions, and mentored engineering students
2004 — 2007

Selected open source

Antaḥkaraṇa — governed, lifelong-learning AI SDK
50+ machine-checked invariants · provable unlearning · governed agent. pip install antahkarana (+ antahkarana-cli) · HF antahkarana-base · two arXiv papers.
PyPI · HF
DPMM — from-scratch Mixture-of-Experts LLM
A research-grade MoE language model built end to end — training pipeline, LoRA adapters, safety, inference.
GitHub
20+ benchmarking frameworks & tools
GPU networking, quantization, speculative decoding, distributed training, LLM serving, observability.
GitHub · HF

Publications

Making AI safety a falsifiable, machine-checkable property of every release.
arXiv:2607.13070
Growing a family of machine-checked invariants (INV-1…52) while preserving earlier guarantees.
arXiv
Reference architectures and technical articles.
blogs.oracle.com

Writing · Beyond the Model

A weekly newsletter where AI infrastructure meets practical engineering — deep dives on GPU clusters, distributed training, inference optimization, benchmarking, and governed AI. 533+ subscribers · 24 editions.

Book

Antaḥkaraṇa: The Inner Instrument — book cover

Antaḥkaraṇa: The Inner Instrument

Part One · eBook · ISBN 9798235023017

Reading a 2,500-year-old map of the mind — the antaḥkaraṇa, the "inner instrument" — as an engineer, and building it organ by organ into a real, measured continual-learning AI, with the failures kept in and every number you can check.

Beyond technology

Away from AI infrastructure, I find balance in cricket and music. Cricket — a lifelong passion — is all strategy, patience, and the balance of individual excellence with team play; the same instincts I bring to complex infrastructure work.

"Music teaches rhythm, timing, and the art of harmonious collaboration — qualities essential for orchestrating complex AI deployments."

Contact

Open to conversations on AI/ML infrastructure, GPU platforms, and governed AI. The best way to reach me is LinkedIn.