Beyond the Model is my weekly newsletter about where AI infrastructure meets real engineering. I write deep dives on GPU clusters, distributed training, inference optimization, benchmarking, and governed AI. 533+ subscribers, 24 editions so far.

Subscribe on LinkedIn

All editions

The Antaḥkaraṇa arc, building a governed AI mind, v6 → v9 (and, in v9, honestly quantum) NEW
♥ 5
1 week ago
Antaḥkaraṇa: An AI Architecture, a gate that proves itself
♥ 11
2 weeks ago
Antaḥkaraṇa now measures itself, the Continual-Governance Benchmark
♥ 6
3 weeks ago
It learnt without forgetting. Now it acts and governs itself, a Vedic model of mind as a research agent that audits real systems
♥ 6
1 month ago
Vedic Antaḥkaraṇa is not just a concept anymore, a 2,500-year-old model of mind, now three AI models you can download
♥ 9
1 month ago
A 2,500-year-old model of mind, rebuilt as working AI, that learns for life and refuses to hallucinate
♥ 8
1 month ago
We replaced a wind tunnel with a neural net. The cheap model won, and we found exactly where it lies
♥ 18
1 month ago
How to Build a Multi-Capability LLM From Scratch, the complete technical blueprint
♥ 12
2 months ago
Which Mixture-of-Experts model should you actually deploy?
♥ 11
2 months ago
The KV-Cache Bake-Off, TRT-LLM vs vLLM on the same reasoning model
♥ 7
3 months ago
Why Your AI Training is Bottlenecked by the Network (Not GPUs), and how to fix it
♥ 19
4 months ago
Why Your GPUs Are Waiting: the hidden network bottleneck in AI training
♥ 17
4 months ago
The Hidden Performance Killers in Your LLM Training Pipeline
♥ 8
5 months ago
The Hidden Cost of Choosing the Wrong Distributed Training Strategy
♥ 23
5 months ago
We Cut Our LLM Inference Costs by 71% Without Sacrificing Quality, here's how
♥ 5
5 months ago
LLM Observability Stack v2.0: why infrastructure testing changed everything
♥ 18
5 months ago
Beyond MMLU, why traditional AI benchmarks are failing us
♥ 4
6 months ago
Understanding Reasoning-First LLMs: a benchmark study of NVIDIA Nemotron-3-Nano
♥ 13
6 months ago
LLM Observability: why monitoring your AI infrastructure is no longer optional
♥ 56
6 months ago
RAG Quality vs Speed: a framework for measuring what matters
♥ 4
6 months ago
Speculative Decoding: get 14-17% faster LLM inference without changing your model
♥ 15
6 months ago
MoE Inference: 7 benchmarks that reveal what actually happens when Mixtral serves tokens
♥ 4
6 months ago
Mixture-of-Experts (MoE) Training Parallelism: scaling expert networks efficiently
♥ 13
6 months ago
LLM Training Parallelism: a practical guide to choosing the right strategy
♥ 8
6 months ago