shreyan basu ray
/home /work /consulting /writing /about
writing ✦ articles, papers, policy

Writing on machine learning, agents and responsible AI.

Technical articles, a book, a peer-reviewed chapter, a working paper and policy work on responsible AI.

17+
technical articles for weights & biases
1
published book, free to read
2
policy documents sent to MeitY and NASSCOM

Technical articles

all on weights & biases →
Agentic workflows in production: mechanisms, trade-offs and design patterns agents Getting started with embodied chain-of-thought (ECoT) embodied ai GLM-5.2: The open coding model that put the critic back into RL reinforcement learning Understanding the Muon optimizer, tested against AdamW with NanoGPT optimizers Fine-tuning Qwen 3.5 on multimodal data with Unsloth and W&B Weave fine-tuning Python tutorial: AI agent observability with W&B Weave observability LLM benchmarking and evaluation: a blueprint for technical leaders benchmarking Understanding and applying the F1 score evaluation Intelligent document processing in finance finance The uses of LLMs in legal: a blueprint for AI leaders legal What is MLOps? Machine learning operations explained mlops Anomaly detection: an introduction with a Python tutorial anomaly detection How to use W&B Inference with Claude Code tooling

Papers, book and policy

working paper ✦ july 2026

Agent Complexity Theory

Choose an agent architecture before you build it, not after benchmarking. Complexity signatures for four machine families with matching lower bounds, reconciled against ten production workflows.

33 pages ✦ four case studies ✦ one documented failure

framework

SafeTrust: agentic development in India ↗

A framework for building and deploying agentic systems responsibly in the Indian context.

book

The Architect's Handbook: Engineering Responsible AI ↗

A practitioner's handbook on building AI systems that can be governed, audited and explained.

Eighty-six pages ✦ version 1.0, january 2026

peer reviewed

Quantum Neural Networks ↗

Where quantum machine learning actually earns its cost against classical baselines.

Springer Nature, with Soujanya Ray ✦ ISEM vol 65, 205 to 233

policy

C-RAI ↗

A certification framework for responsible AI, proposing how Indian AI systems could be audited and labelled before deployment.

Sent to MeitY and NASSCOM ✦ draft v2, 2026

Want a piece written on your stack?

I write from experiments I run myself, which means I need access to the system rather than a brief about it.

Get in touch → Read on W&B
home github linkedin x scholar weights & biases npm events whatsapp email
reviewed august 2026