A program of CascadAI

GAMBIT

Gamified AI Bots for Interactive Training

Students walk into a shared virtual campus and meet a tutor in every room — one per unit of the course, grounded in that unit's real material. They ask out loud, build something, and are assessed on the artifacts they produce rather than on prose a chatbot could have written for them.

See how it works Partner with us

Running now in two live courses · hosted and operated by CascadAI · built in Victoria, BC

gambit.cascadai.org — campus view
One room per unit. Each door is a bot that knows that unit's readings, protocols and deadlines.

Three problems that arrived at once

Generative AI did not create the crisis in teaching on its own — it collided with two older ones. GAMBIT is built against all three together.

The take-home assignment stopped working

Detection is unreliable and prohibition teaches nothing. The way out is assessment that watches the work happen — questions asked, sources chosen, artifacts built — instead of grading a finished text.

Tutoring works. It has never scaled.

One-to-one attention is among the strongest interventions education has, and the one almost no public institution can afford to give. A tutor per unit, always awake, is the first honest shot at closing that gap.

Online courses are lonely places

Video lectures deliver content and lose people. What keeps a cohort together is the sense of being somewhere with others — seeing a classmate cross the room and being able to stop them.

What GAMBIT is

A campus you can walk around, with a specialist behind every door

The course map is the interface. Each room is one unit; standing in it puts you in voice and text range of that unit's bot and of everyone else in the room. Course materials sit in a panel beside the conversation, so the answer and its source are on screen together.

A GAMBIT room in CSC 110: two avatars standing on a highlighted floor tile marked ask Dr. Berg, with the official Fall 2026 course outline and its section schedule open in a panel on the right.
CSC 110, Fundamentals of Programming I — stepping onto the marked tile opens a conversation with Dr. Berg, the unit's tutor. The official course outline stays open beside it.
01

A place, not a page

Learners appear on a shared map as avatars. Proximity opens voice and video, so office hours, study groups and overheard questions happen the way they do in a building.

02

One tutor per unit

Every bot is scoped to its own unit and briefed on that unit's readings, protocols and assessment. A narrow tutor that knows the syllabus beats a general assistant that knows everything else.

03

Sources, not just answers

Ask for the literature and you get the papers, listed and linkable in the panel beside the chat. Students learn to check the tutor — which is the skill the whole thing is for.

Grounded, not guessed

Every claim comes back with its paper attached

Ask a GAMBIT tutor for the literature on a topic and it searches the real corpus, answers in plain language, and files what it cited into a “papers cited in this room” list that the whole cohort can see and follow.

The citation list is shared, so a room accumulates a reading list built by the questions the students actually asked.

A GAMBIT room where a student asks the Rosalind tutor for papers on phages in B. subtilis; the reply lists specific preprints and a panel beside it shows the papers cited in this room.
The progression

You level up by producing research, not by answering quizzes

Each learner carries a character sheet. An idea starts as one line — use a toehold sensor to detect Zika — and advances through eight slots as the evidence behind it accumulates: references, constructs, protocols, results, papers, patents, presentations.

The bot behaves like a supervisor rather than an answer machine. It will not fill a pocket for you; it asks what you have, tells you what is missing, and points at where to look. Progress is the artifact trail, which is also exactly what an instructor needs to grade.

  • Assessment evidence is generated as a by-product of the work, per learner, per idea.
  • Ideas can be marked complete or abandoned — dead ends are recorded, not hidden.
  • Learners make their own avatar, so the person in the room is recognisably them.
In the field

Two courses are running on it today

One graduate-level bioengineering course and one first-year programming course — deliberately far apart, to find out what is general about the model and what is not.

Graduate

How To Grow Almost Anything

A synthetic-biology course whose modules run from cell-free systems to mammalian manufacturing. Each module room holds a Rosalind tutor with retrieval over the preprint literature, so students chasing a design question get real papers back and a shared citation list per room.

Ten module rooms plus a final-projects hall.

First yearFall 2026

CSC 110 — Fundamentals of Programming I

A large multi-section intro course, the hardest possible test: many lecture sections, a wide spread of prior experience, and the exact population most exposed to quietly outsourcing the work. Eleven units, one room each, all tied to the official course outline.

Dr. Berg answers in every unit room, from the same outline students were given.

Open roots, run by us

GAMBIT is built on open-source foundations and hosted and operated by CascadAI for each partner course. Releasing the platform for others to run themselves is where funding takes it next.

A campus authored per course

We build the map with the instructor and version it like any other courseware.

WebRTC voice and video

Proximity-based, in the browser, nothing to install.

Retrieval-grounded tutors

Each bot is scoped to its unit's corpus and cites what it used.

Hosted where the institution needs it

Canadian research cloud or AWS, chosen with the partner's data requirements.

Who we work with

Built with institutions, not sold to them

GAMBIT exists because classrooms, a research network and a compute provider each carried part of it.

University of Victoria

Northeastern University

Vancouver · Seattle · Oakland

AI Network BC

Pacific Coastal Computing

AWS

Your institution

We are taking on a small number of courses for the next intake, hosted and supported by our team.

What support pays for

CascadAI is a nonprofit. GAMBIT is funded by grants and philanthropy so that it can stay free to the courses that need it most.

More classrooms

Authoring a course map, briefing its tutors and supporting the instructor through a first term.

Finding out if it works

Evaluation designed with our university partners, published either way — including the parts that do not work.

Hardening it for release

Engineering the platform past pilot quality, so that after this round other institutions can run it themselves and the code can be released.

Bring GAMBIT to your course

Tell us the course, the size of the cohort and what you are trying to fix. We will tell you honestly whether this is the right tool for it.

hello@cascadai.org Victoria, BC · replies within a week