Research earns its keep by shipping
We do not maintain a wall between the lab and the product. A result that cannot survive real users, real latency and real edge cases is an unfinished result.
Eullar Labs · applied AI research · Accra
AI that has towork on aTuesday morning.
Eullar Labs builds practical AI tools and publishes what it learns making them. Two are in the field today — Syllabi for teaching and learning, and Reevue for applicant feedback.
- products in field
- 02
- decision authority
- human
- model layer
- swappable
- origin
- 5.60°N 0.19°W
prerequisite graph extraction✦learner state estimation✦evidence-anchored critique✦uncertainty under latency budget✦cohort skew monitoring✦curriculum path rewriting✦criterion ↔ passage alignment✦delayed-recall evaluation✦human override surfaces✦retrieval over syllabi✦prerequisite graph extraction✦learner state estimation✦evidence-anchored critique✦uncertainty under latency budget✦cohort skew monitoring✦curriculum path rewriting✦criterion ↔ passage alignment✦delayed-recall evaluation✦human override surfaces✦retrieval over syllabi✦prerequisite graph extraction✦learner state estimation✦evidence-anchored critique✦uncertainty under latency budget✦cohort skew monitoring✦curriculum path rewriting✦criterion ↔ passage alignment✦delayed-recall evaluation✦human override surfaces✦retrieval over syllabi✦prerequisite graph extraction✦learner state estimation✦evidence-anchored critique✦uncertainty under latency budget✦cohort skew monitoring✦curriculum path rewriting✦criterion ↔ passage alignment✦delayed-recall evaluation✦human override surfaces✦retrieval over syllabi✦
What we are
Alabwithaproduct,notaproductwithablog.
MostofwhatanAIsystemgetswrongshowsuponlywhenarealpersondependsonit—ateacherplanningtomorrow'slesson,agraduatereadingarejection.Sowedonotkeepawallbetweenresearchanddeployment.Thesamepeoplewritetheevaluationandanswerthesupportticket.
We work on two problems where a better system changes an outcome rather than a metric: what a learner is taught next, and what an applicant is told.
eullar · eval
›▍
fig. 0 — an evaluation run, edited for length
ProductsTwo surfaces.
Two surfaces.
Try them right here.
These are not screenshots. Switch profiles, hover the graph, toggle the engine off — the figures below run the same logic the products do, on fixed sample data.
Teaching and learning, tailored to the learner in front of you.
Open Syllabifig. 1 — prerequisite graph, live re-plan
Year 9 · stalls on simultaneous equations
Blocking concept
Proportion
Failures on simultaneous equations trace back through graphing to an unstable proportion model — not to the topic being assessed.
Generated route — 4 steps
- 1Repair. Proportion as a multiplicative relation, using the ratio work already secure
- 2Rebuild. Linear equations re-derived from the repaired proportion idea
- 3Bridge. Graphing lines as the visual form of the same relation
- 4Reach. Simultaneous equations, introduced graphically before algebraically
ArchitectureOne stack,
One stack,
two surfaces.
0
products in the field
Syllabi and Reevue, both running against real cohorts rather than benchmarks.
0pts
recall gained at 14 days
Graph re-planning against topic order, in the evaluation run on the right of this page.
0%
outcomes decided by people
Neither product ranks or selects. They explain, propose and draft; a person decides.
How we work
Six commitments we can be held to.
Written to be falsifiable. If we break one, it should be obvious from the outside.
keep scrolling
The human keeps the decision
Our systems explain, propose and draft. Teachers keep authority over a learning plan; hiring teams keep authority over a hiring outcome. We build the seams that make override easy.
Specificity over reassurance
Vague output is a way of hiding uncertainty. We would rather a system name a narrow gap it can defend than produce a fluent answer to a question it did not understand.
Measure the thing, not the proxy
Satisfaction scores, engagement time and completion rates are proxies. We instrument for retained understanding and acted-on feedback, even when they are slower and less flattering.
Publish the failures
Our working notes include the methods that did not hold up. A lab that only publishes wins is a marketing department with a LaTeX template.
Build where it is needed
We are based in Accra and build first for classrooms and job markets that most AI products treat as an afterthought. Constraint is a design input, not an excuse.
Research
All notesWorking notes, including the ones that failed.
Work with the lab
Ifyourunaclassroomorahiringqueue,wewantthehardversionofyourproblem.
Pilots, research collaborations and integration work. Tell us what breaks today and we will tell you honestly whether we can help.