SR3H is an independent AI research and software development lab in Oxford.
We investigate how AI works and build useful software. Our research focuses on improving the quality and relevance of AI output while protecting personal data and keeping people in control.
Based at Grassroots, the science and technology workspace managed by Oxford Innovation Space.
Personal profile and AI agent
Your profile and personal AI agent, with control over what is shared. MyLegend introduces you, answers questions from information you approve, and lets people request a reply. Explore ↗
Research project · Context and privacy
Personal context for AI, with control over what is shared. Brane explores how selected information can improve an answer without exposing an entire personal history.
Its Safari extension also shows which trackers are present on each page.
Public website and MCP app
Clear guide-price ranges for UK building work, showing the assumptions and cost drivers behind each estimate. Also available as an MCP app for AI assistants. Website ↗
In development · Software engineering
A software development workspace that keeps requirements, decisions, code and tests connected, so changes can be reviewed against what the product is meant to do. Explore ↗
In-house research tool
AI visibility and analytics software for testing how AI finds, describes and recommends businesses. Signal runs batches of questions, records answers and sources, and compares results across repeated tests.
Private workspace · Briefings
A private workspace for people preparing briefings from complex source material, with each person’s work kept separate.
Navigation project · Public website
Navigation that preserves the rider’s chosen route and makes rerouting an explicit decision. Website ↗
We build prototypes to test ideas, compare approaches and find out what is worth developing further. Our work covers how AI systems use information, how people work with them and what makes their results useful.
We investigate what makes an AI response accurate, useful and relevant to the task. This includes how systems use context, draw on sources and retain information, and how their output can be checked.
We research how to give AI useful context without handing over an entire personal history. This includes on-device processing, selective sharing and clear choices about what is remembered, used or sent to an external service.
We develop ways for AI to work with software and practical tasks. We test where automation helps, where fixed rules work better and where a person needs to review or approve the result.
We study how AI finds information, compares options and recommends businesses, products or services. We examine the evidence behind those recommendations and the role of commercial influence. Signal is one project exploring these questions.
We welcome conversations about research, product design and software development. Tell us what you are trying to understand or build, and where you need help.
hello@sr3h.uk
