WizzStudy
Get started
Research

Faculty Academic Supervisor — Mitacs Accelerate Partnership

Host industry partner: 9keys Group Inc. (WizzStudy) Funding mechanism: Mitacs Accelerate ($15k per intern term, no faculty time cost) Commitment: ~3 hours / week during the term; co-author credit on resulting publications Eligibility: Tenure-track or tenured faculty at a Canadian post-secondary institution

Why this exists

Mitacs Accelerate is structured for an industry partner to fund a graduate student's research term, with academic oversight from a faculty supervisor at a Canadian university. The faculty supervisor:

  • Validates that the research is genuinely thesis-grade.
  • Signs off on the project plan and the final report.
  • Co-mentors the intern alongside the industry team.
  • Co-authors any resulting publications.

There is no cost to the faculty member's lab budget. Mitacs and 9keys cover the intern stipend in full.

The research program

We are studying summarization faithfulness in source-grounded study tools and its causal relationship to long-term retention in student users. The product (WizzStudy) is a deployed system with ~thousands of users generating real review data, instrumented end-to-end:

  • Every generation logs a prompt_version tag, token usage, latency, cost, faithfulness score.
  • Every flashcard review logs an FSRS rating and an IRT-fit difficulty.
  • A nightly eval harness re-scores prompts against a hand-graded corpus.

This is a rare combination for academic research — a real-world deployment, instrumented at the row level, with IRB-style consent already in place (research opt-in defaults are user-controllable).

Open research questions

  1. Faithfulness operationalization. ROUGE-L and BERTScore reward surface overlap. LLM judges have known bias. What's the right metric — or family of metrics — for citation-grounded summary fidelity? (Submission targets: TACL, EMNLP Findings.)
  1. Retention attribution. Given a card generated by variant V, exposed to student S, with FSRS history H, can we attribute long-term retention deltas to V vs. confounds (S's prior knowledge, exposure schedule)? Causal-inference techniques on observational deployment data. (Submission targets: L@S, LAK, EDM.)
  1. Citation-first generation as a learning intervention. Does seeing an inline citation while reviewing change learning outcomes vs. an uncited claim? Two-arm field study, RCT in production. (Submission targets: AIED, L@S.)

The intern picks one (with your input). You co-supervise. The team handles the engineering.

What the partnership involves

  • Weekly check-in. 30–60 min with the intern. We're happy to attend or stay out.
  • Project sign-off. You approve the project plan at the start, and the final report at the end. Mitacs provides templates.
  • REB / ethics review. We have an REB application skeleton (docs/research/reb-application.md) ready to file at your institution. The intern leads the filing; you sign as PI.
  • Publication. Co-author credit, in author-order to be agreed up front. We do not publish anything you haven't seen.
  • IP. Mitacs Accelerate allows up to 50% IP retention for the academic side. We're flexible and have no interest in restricting the intern's ability to publish the dataset, the harness, or the methodology under open licenses.

What we ask of you

  1. A 30-minute introductory conversation. We share our actual eval reports, code, and consent flow.
  2. If you're interested, you nominate (or we co-recruit) a graduate student.
  3. You sign the Mitacs Accelerate proposal as academic supervisor.
  4. We're in business.

Contact

Anass Elabeidi — Founder, 9keys Group / WizzStudy research@wizzstudy.com wizzstudy.com/research

Documents we can share on request, before any commitment: