About

A small studio for the unglamorous half of AI work.

PromptSri.AI started because the same problem kept appearing: teams with capable models, impressive demos, and no way to tell whether last week's prompt change made anything better. We build that missing layer.

What we believe

A number or it did not happen

Every recommendation we make comes with a before and after on a fixed eval set. If we cannot measure the improvement, we do not claim it.

Prompts belong in the repository

Not in a dashboard text box that nobody can diff. Versioned, reviewed, and logged with every response they produce.

The model is not the author

It removes friction from research and structure. Judgement, point of view, and accountability stay with people.

Handover is the deliverable

We are a bad long-term dependency by design. You keep the harness, the library, and the runbook.

Who you work with

Sri Raman

Prompt systems

Spent six years on search relevance before language models made prompt design a discipline. Writes the audits.

Ananya Iyer

Retrieval & evaluation

ML engineer. Builds the eval harnesses and argues, correctly, that retrieval metrics come before prompt tuning.

Meera Nathan

Content systems

Editor turned pipeline designer. Keeps the model out of the parts of writing that require having a point of view.

Work with us

We take on three engagements at a time, which keeps the work honest and the calendar sane.

Start a conversation