Inspiration

Evove Academy supports fitness education in the Maldives, where learners, tutors, and administrators all need timely guidance across attendance, payments, assessments, coursework, and learner progress. A learning management system can store this information, but staff still have to interpret it, identify risks, and decide what action to take next.

We built the AI Coach Layer to make that operational knowledge easier to use. The goal is not to replace tutors or admins, but to give them faster, clearer, evidence-based support so learners receive help earlier and staff can focus on human decisions.

What it does

The AI Coach Layer adds Gemini-powered intelligence to the Evove Academy LMS. It helps staff understand learner progress, spot support needs, draft interventions, and prepare operational summaries using existing academy data.

Key capabilities include:

  • Admin learner success insights and risk summaries
  • Cohort-level operational briefs
  • Tutor-facing draft feedback and follow-up suggestions
  • Learner guidance for next steps and study preparation
  • Revenue and business evidence explanations for admin review
  • AI audit tracking so usage, edits, failures, and outcomes remain measurable

The system is intentionally designed as a decision-support layer. AI-generated messages are drafts only, and staff must review and send them through the existing LMS workflow.

How we built it

We built the project into Evove Academy’s existing full-stack LMS. The backend is a Laravel API, the frontend is a Next.js portal, and the system runs on Google Cloud. Gemini is used as the primary generation model for learner success, tutor support, and business evidence workflows.

A major design principle was context minimization. The AI layer receives only the operational details needed for the task, such as cohort, study option, progress, attendance, payment status, pending assessments, and upcoming sessions. It avoids sending unnecessary personal identifiers or private documents to model providers.

We also added auditability into the workflow. AI interactions can be tracked by actor, role, portal, provider usage, failures, and reviewed outcomes, making the layer safer and easier to improve over time.

Challenges

The hardest part was making the AI useful without making it too autonomous. In an education and finance-sensitive environment, the AI must not change attendance, marks, invoices, learner access, or messages on its own.

Another challenge was designing prompts and workflows that work with real LMS context while staying privacy-conscious. We had to separate admin, tutor, and learner use cases clearly, and make sure each role only receives guidance appropriate to their responsibilities.

What we learned

We learned that AI is most valuable in an LMS when it is connected to real operational evidence, not used as a generic chatbot. The most useful outputs are specific, reviewable, and tied to existing workflows.

We also learned that safety is a product feature. Clear role boundaries, minimized context, draft-only actions, cooldowns, and audit logs make the AI layer more trustworthy for staff and learners.

What's next

Next, we want to improve the quality of cohort-level insights, expand outcome tracking, and make it easier for administrators to measure whether AI-assisted interventions lead to better learner progress, faster follow-up, and clearer business evidence.

Live Product and Evidence

The production LMS is live at evoveacademy.com. Because it contains real learner, finance, and academy data, we are not providing open demo credentials.

Instead, the submission includes a private evidence repository and a sanitized judge evidence page with screenshots, implementation notes, running evidence, financial evidence, and AI workflow proof. This lets judges review the product and business evidence without exposing learner records, payment data, private documents, or internal operational comments.

Built With

  • cloud-sql-for-mysql
  • docker
  • gemini
  • gemini-api
  • github-actions
  • google-cloud
  • google-cloud-run
  • laravel-12
  • laravel-sanctum
  • mysql
  • next.js
  • nvidia-nemotron
  • php-8.2
  • phpunit
  • playwright
  • react-19
  • rest-api
  • sentry
  • tailwind-css-v4
  • typescript
Share this project:

Updates