Notes from building private AI.
Notes from building private AI for regulated teams: engineering, evaluation, deployment and the rules around them.
Featured · Trust · 5 November 2026 · 8 min read · By Arthiq
What a bank’s vendor review asks an AI supplier.
The documents, questionnaires and contract terms third-party risk teams request from an AI supplier, and how deploying inside the bank’s environment changes the review.
Read the article →Delivery · 19 October 2026 · 5 min read · By Rajesh Raachabattuni
You have the code. That is not the same as owning the system.
A handover gives you the repository, a diagram and a spreadsheet of logins. Ownership is something else: whether a change can reach production without the person who built it. Here is how to find out before you need to.
Deployment · 16 October 2026 · 6 min read · By Rajesh Raachabattuni
The build is a one-off. An AI feature charges you every time it works.
Conventional software costs money to build and almost nothing to run. An AI feature inverts that, and the bill grows exactly as adoption does — so the number that decides whether it is worth having is the cost per case, not the cost of the build.
Evaluation · 12 October 2026 · 6 min read · By Rajesh Raachabattuni
Anyone who can send you a document can talk to your agent.
Most controls treat prompt injection as a bug waiting for a patch. It follows from how these models read, which means the only defence that holds is designing so that a successful injection is not worth anything.
Evaluation · 2 August 2026 · 5 min read · By Rajesh Raachabattuni
You can’t unit-test an agent. So how do you know it works?
Normal software fails loudly and repeatably. Agents fail quietly, differently each time, and only on the cases you didn’t imagine. Here’s what testing one actually looks like.
Applications · 17 July 2026 · 5 min read · By Rajesh Raachabattuni
The hard part of an AI agent isn’t the model. It’s the leash.
Every agent demo I’ve built worked because the agent could do anything. Every one that reached production worked because I’d spent most of my time deciding what it wasn’t allowed to do.