Designed with a public university in Ghana
Your QA office already collects evaluations. OneRubric turns them into verified improvements by next term.
Students rate and comment. A private AI reads every comment. Lecturers commit to action plans. The next period’s data confirms whether the score moved. One PDF proves it to your council.
7-day trial on the hosted service · Institutional SSO only · Self-hosted option
What OneRubric replaces
Excel and manual aggregation
Today · Stores responses
With OneRubric · Aggregate analytics and an AI summary from the same data
Google Forms and Sheets
Today · Collects evaluations
With OneRubric · Action plans and automatic next-term verification
SIS / ERP evaluation modules
Today · Heavy, LMS-tied modules
With OneRubric · Stand-alone and lightweight; no ERP integration required
Imported US survey suites
Today · Mature analytics
With OneRubric · African institutional terminology and a market-appropriate model
Closing the loop
From what students said to what verifiably changed
Most evaluation tools stop at a score. OneRubric tracks what happens next, across periods, and reports the result in one line.
- 01
Expectations, then evaluation
At the start of term students record what they hope to get from a course. At the end they rate it on a five-point scale and comment anonymously. Progress saves as they go.
- 02
Every comment is read
A private AI reads 100% of comments, surfaces themes, and answers questions with the actual anonymous comments as cited evidence. A cohort-relative anomaly check flags sharp drops.
- 03
A plan with a baseline
When an aspect is flagged, the lecturer commits to concrete action items. The system records the aspect’s score as a fixed baseline. The head of department reviews and approves.
- 04
Verified next term
A scheduled job re-measures the same aspect in the next period and stamps the verdict: improved (+0.20 or more), regressed (−0.20 or less), or no change. Leadership sees it in aggregate.
“83% of flagged aspects have an action plan; 60% have verifiably improved by the next period.”
Product tour
Seven roles. Each sees exactly their slice.
Evaluation is a chain of responsibility. Access is scoped per user, so a head sees their department, a dean their school, and the executive only aggregates.
The control centre for the whole quality cycle
Period lifecycle, people and access, integrity cases, anomaly detection, branding and the audit log — everything the QA office runs, in one place.
- Create, open, auto-close and archive evaluation periods; bulk-load rosters
- Investigations with timelines, evidence attachments and a recorded verdict
- Review and quarantine queues for flagged comments; humans decide
- Cohort-relative anomaly detection: a hard semester does not trigger false alarms
- Append-only audit log of every privileged action, exportable
- Anonymity threshold, custom questions and branding set from admin screens
Private AI
AI that stays on your infrastructure
Student comments are sensitive. The AI layer is built so that no student identifier ever crosses the AI boundary, whichever way you deploy.
Self-hosted: Mistral 7B via Ollama, on your servers
The model runs on a VM you control, on a private network, with no outbound AI traffic. OneRubric the company has no sub-processors in this model.
Hosted service: a contracted inference provider
On the SaaS edition, inference runs at Together.ai by default (OpenAI optional, off by default) under a DPA with zero-day prompt retention and a contractual bar on training. Only anonymised, aggregate text and ratings are sent.
- Answers cite the actual anonymous comments, with similarity scores. The assistant never invents a quote.
- Retrieval respects each role’s scope: a lecturer’s “self” scope cannot surface another lecturer’s comments.
- The chat model is swappable from the operator UI with no restart; the embedding model is guarded against misconfiguration.
- Token usage is metered per month against a quota with an 80% warning, so costs stay predictable.
- When the model is offline the surface shows “AI unavailable” instead of breaking.
Trust
Anonymity is structural, not a setting
Students have to believe their feedback is anonymous and lecturers have to believe the process is fair. Both are enforced in the data model, not just promised in a policy.
Anonymity floor
Course aggregates and comments are withheld until a configurable number of distinct responses is reached, so one student in a small class cannot be re-identified.
No identity on written reviews
A review row is comment and rating only. Reports and AI surfaces never join a comment back to a student email; the AI never receives that column.
Disputes and investigations
A lecturer can contest a comment that breaks the rules. An upheld dispute hides it, removes it from the leaderboard and deletes its AI embedding in one step. Serious matters become case files with evidence and a verdict.
Append-only audit log
Every privileged action is recorded with actor, action, target and timestamp. There is no UI to edit or delete it, and it exports to CSV for procurement or accreditation.
Institutional SSO only
Google Workspace, Microsoft Entra (single-tenant only), SAML 2.0 with signed assertions and replay protection, and SCIM 2.0 provisioning. No passwords, by design.
Humans decide
The misconduct scanner is a conservative keyword matcher that only surfaces candidates for review. Dismissing a false positive quarantines similar comments for a person; nothing is silently cleared.
Built with a QA directorate
Designed in a real evaluation cycle
OneRubric was designed alongside the quality-assurance directorate of a public university in Ghana, from the way its evaluation cycle actually ran. It is now open to the first cohort of institutions.
The starting point
Evaluations collected every semester, thousands of comments nobody has time to read, and a period report the QA office writes by hand over about two weeks.
What the partnership shaped
The seven-role hierarchy, the six-area rating structure, custom questions per programme, the dispute process, and white-label branding from admin screens.
On their terms
Runs on infrastructure the institution controls, wears its branding, uses its existing Google and Microsoft sign-in, and keeps student comments inside the institution.
Join the first cohort
We are onboarding a small number of institutions now. Early access, direct input on the roadmap, and a named engineering contact for your QA office and IT team.
Plans
Three ways to run it. Same software.
Pricing will be announced at launch. Until then, talk to us: we will tell you plainly what each model involves for an institution your size.
Hosted
Multi-tenant service operated by OneRubric at your-institution.onerubric.click.
Pricing announced at launch
- Live after a short setup wizard; 7-day trial
- Data region pinned at provisioning (US, DE, UK, IN or SG) and never moved without 30 days’ notice
- Daily snapshots, 7-day point-in-time recovery
- Hosted AI under a no-training, zero-retention DPA
- Google, Microsoft and SAML sign-in
Self-hosted
Single-tenant deployment on infrastructure your IT team controls.
Pricing announced at launch
- Your database, your VMs, your region; no OneRubric sub-processors
- AI on your own Ollama server; no outbound AI traffic
- Deploys to DigitalOcean, AWS, GCP, Azure or on-premise
- Source access and a customer-commissioned audit per contract year
- You decide when to upgrade; there is no kill switch
Managed deployment
For institutions without spare IT capacity: we operate OneRubric inside your own cloud account.
Pricing announced at launch
- You own the account, the data and the infrastructure
- We run the application, Postgres and AI VMs as a service
- The same self-hosted privacy posture
- Shared-responsibility split agreed in the contract
Customer-initiated full data export is always free and never tied to contract status.
FAQ
Common questions
Can lecturers or administrators see who wrote a comment?
Where does the AI run, and does it learn from our data?
What does “verified next term” actually mean?
How do people sign in?
Can we self-host instead of using the hosted service?
What happens to our data if we leave?
Do you hold SOC 2 or ISO 27001?
How long does it take to get started?
What does it cost?
See it on a sample institution
A live walkthrough with your questions first, or a 7-day trial on the hosted service with no card required.
Procurement or IT questions first? Read the security overview.