← Selected work

Continuity / AI systems

What should an AI
carry forward?

An independent project exploring how AI applications can keep useful context across conversations.

Focus
AI applications & operational visibility
Built with
Python · Next.js
Status
Implemented components / ongoing integration
A conversation ends. Useful context should not have to start over.
Session 01Useful contextSession 02

Exploring what it takes to make
the next conversation more useful.

Conceptual illustration

The question behind the project.

When working with AI applications, I kept returning to the same question: what happens to useful context when a conversation ends or the model changes?

I began building and testing components to explore that problem. The work combines AI application development with tools for understanding the system’s operational state.

A glimpse of the product

Making system status easier to read.

An operations view brings observations into one place. This recreated preview uses fictional workspaces and sample states, with no connection to the private system.

Continuity / OperationsSynthetic demoContext deserves
a little continuity.
Know what was observed. Know what remains unknown.Workspace AFreshRecent observationThread healthHealthyRetrieval integrityIndex verifiedReader coverageBoundedFull historyUnknownWorkspace BStaleLast known · not liveThread healthLast known healthyRetrieval integrityLast known verifiedReader coverageBoundedFull historyUnknownWorkspace CUnknownNo observation availableThread healthUnknownRetrieval integrityUnknownReader coverageUnknownFull historyUnknownRead-only observations No conversation content

Recreated dashboard · Synthetic workspaces and states · No connection to the private system

My contribution.

I own the product requirements, architecture decisions, and acceptance criteria. AI-assisted tools supported implementation.

My work spans defining the problem, making technical decisions, and evaluating the implemented components. The underlying implementation remains private.

Where the work stands.

Memory and operations components have been implemented and tested. Integration work and the related AI application remain in development.

The tests cover specific component behaviours. They do not establish end-to-end reliability or scientific validation, and this is not presented as a finished consumer product.

Another part of the story

TATAA different kind of context: running a family business.