Your lab manual becomes the mentor, and your rubric becomes the marker.
Evaluation checks
0/4 ruled
Attainment so far
0 of 100 weight settled
0/100
The surface that runs every week, not once a year.
You bring the assignment you already set. We index the manual, the SDK docs and the reference repos into a knowledge graph the mentor agent draws on to answer students' questions well, and every submitted repo comes back scored against your rubric, not ours.
How it runs
The practical is written down, and the bar with it, before the session starts.
You write the problem statement
The practical as you already set it: what lands, what has to come back, in the words a team lead would use. We do not rewrite your lab.
Deliverables, stated rather than implied
Each one an artefact a marker can point at — the handler, the totals, the file written back, the test. Students read the same list you mark against.
Evaluation criteria, weighted to a hundred
Each criterion carries its weight, where in the code the answer should live, and what separates a pass from a fail. They publish with the sheet, so the bar is stated rather than guessed at.
Marks alone do not say what a student could and could not do, so someone ends up rebuilding that from memory months later. Here every check is recorded as the work is marked, so the record is written while it is still true.
One line per check
You get a result for each check you wrote, not one mark for the whole lab. It is clear which part a student got right and which they did not.
Every score points at the code
Each result names the line that earned it. Anyone reviewing the work later can open the repository and see for themselves.
Yours as a spreadsheet
One file per semester with every submission, check and result. Download it, keep it, share it with whoever asks.
These titles barely existed when most syllabi were written. What each one needs is teachable, but only where students integrate, direct agents and ship.
designs the system that runs itself
Orchestrates systems where several agents carry a complex, multi-step workflow autonomously. Defines long-term memory, planning logic, decision constraints and tool access, and stops infinite reasoning loops and goal drift before they reach production.
puts the model in the product
Bridges product engineering and model integration, embedding foundation models into applications that have to scale. Designs retrieval pipelines, manages embeddings, fine-tunes task-specific models, and keeps latency and token cost from eating the margin.
keeps non-determinism honest
Makes sure model output, agents and non-deterministic logic stay accurate and within spec. Builds automated evals, benchmarks model drift, guards against prompt injection, and watches hallucination rates in production rather than in a demo.
replaces the frontend / backend split
Ships whole features rather than one layer of them, from interface to backend logic, working inside AI-first execution environments. The split between frontend and backend roles is what this one collapses.
sits with the customer
Reads an unfamiliar SDK fast, integrates against someone else's API, and ships under a customer's constraints rather than ideal ones. Built by labs set on a real SDK, judged on the integration rather than the output.
owns it after it ships
Wires services together end to end, makes something that runs for somebody else, and decides the trade-offs without a committee. Built by challenges and hackathons, where a repo that works beats a slide that claims it.
catches the agent when it is wrong
Directs an agent to a useful result, reviews code they did not write, and spots a confidently wrong answer before it ships. Built by the mentor in the student's editor, plus a judge that catches confident nonsense.
No curriculum committee, no migration, nothing to rip out. A pilot is small enough for one faculty member to run and honest enough to judge on the results.
You bring
We set up
You leave with
University Connect
One platform for the labs your departments already run and the challenges your students already want — with the rubric, and the evidence it produces, staying yours.
Talk to us
Start where it suits you
One course, one department or the whole institution. We set up the graph, the checks and the access.
Book a walkthroughLook first
See what a track costs
The eight role tracks, the thirty hours inside one, and what your department is left holding at the end of it.
Browse the tracks