What I study,
and what I found.
Work is slower than it has to be, and most of the lost time hides in the gaps between people, tools and decisions. Every venture starts as a question about one of those gaps. I measure it first and publish what I find, including what doesn't work yet.
- questions
- 4
- published
- 4
- in progress
- 3
In progress, and soon out.
How I do research.
- 01
Measure before claiming.
Every number traces back to a run or a public source, with the dataset, version and date beside it.
- 02
Publish the weak results too.
Where something falls short, it stays on the page as an open problem instead of disappearing.
- 03
Compare fairly.
Other systems are scored the way they report themselves, and only numbers we could check at the source are shown.
- 04
Then build the fix.
Each question belongs to a venture. The research says where time is lost; the venture is the attempt to win it back.
- question 01 · memory
How does a team remember what changed?
In a public memory test, AUVY finds the right conversation 99.4% of the time. The places it still falls short are published next to it.
the venture it feeds · AUVY ↗- studyauvy.ai/research · 27 sep 2026How AUVY remembers and finds Memory built the way nervous systems work: what AUVY takes from brain research, its results on public benchmarks beside every system we could check, and the open problems.
- in progress · paperAUVY research · coming soonMemContinuity: does a company's memory stay true over time?Memory tests today ask whether one person's chat history can be recalled. This paper asks the company question: across months of documents, mail, meetings and decisions, does the memory keep up with every update, keep the timeline straight, catch silent contradictions, and stay inside who may see what? A benchmark and a simulated company to run it on.
- in progress · protocolAUVY research · coming soonFrom noticing to actingHow a system should notice that something changed, check it against the source, and get the next step ready for a person to approve.
- question 02 · waiting
Where does the waiting actually happen?
Twenty of 472 rail segments cause 30% of all the extra delay on German long-distance trains.
the venture it feeds · ImTakt ↗- reportimtakt.dev · sep 2026Verspätungen im Fernverkehr 2026 321 million observed stops at 5,457 stations: where delay builds up along a journey, which segments cause most of it, and seven measures Deutsche Bahn and the states could put in place in 2027.
- in progress · studyImTakt research · coming soonHow Deutsche Bahn could run on timeThe full study behind the report. Where German long-distance rail loses its time, from timetable reserves and missed connections to construction, what fixing each would win back for travellers, and a pilot on two corridors to prove it.
- question 03 · trust
How can people and their AI assistants share work without sharing secrets?
Messages are sealed on your own device. We carry them, but we can't read them.
the venture it feeds · OpenPulse ↗ - question 04 · care
What happens to therapy between the sessions?
A week has 10,080 minutes. The session is 50 of them. Limbova is built for the rest.
the venture it feeds · Limbova ↗
Notes from the work, as it happens.
The first notes are being written. They'll appear here, and in the rss feed.