Student Companion AI Platform

The Student Companion AI is an AI-powered platform designed to provide seamless academic, administrative, and personal support to students across higher educational institutions in Rwanda.

Comment

Inspiration

In many African universities, students experience delays and fragmented access to academic and administrative support. Without a centralized solution, student needs go unmet and service delivery remains inefficient, resulting in information inequality, disengagement, and missed opportunities - while institutions lack scalable, well-governed, and ethical AI solutions to address these gaps.

At African Leadership University, we kept seeing the same gap: students have urgent questions — about academic, administration, opportunities, where to get support - and the answers exist, but they're buried and there's rarely anyone to ask after working hours. Meanwhile, the people running student programs have no real-time view of what students are struggling with. We wanted to solve both sides at once: give students an honest, always-on companion, and give institutions a window into student needs — packaged as something a university can actually buy.

What it does

Student Companion AI is a mobile-first, AI-powered peer assistant for university students. Students ask anything about academics, campus life, policies, or opportunities and get a grounded, cited answer in a warm peer voice — built to admit what it doesn't know rather than hallucinate. It also surfaces fellowships, scholarships, and jobs directly in the app.

For the institution, we added a multi-tenant B2B admin console: a live dashboard of student engagement (active users, conversation volume, the top questions students are actually asking), plus tools to manage the opportunities feed, curators and office-hours bookings, and human-escalation inquiries. When an admin publishes an opportunity, it appears instantly in every student's app — no code, no deploy.

How we built it

The frontend is a React + TypeScript + Vite PWA, deployed on Vercel; the admin dashboard was built with v0. The backend is a Dockerised FastAPI service that runs the RAG pipeline — vector search over ChromaDB using sentence-transformers embeddings, answered by Anthropic's Claude with two-block prompt caching to cut cost ~10x. Knowledge is authored by staff in a Google Sheet that syncs into the vector index — no engineer in the loop.

For this hackathon we added the structured-data layer the product was missing: an Amazon Aurora PostgreSQL database that's the system of record for organizations, students, opportunities, curators, bookings, conversations, and inquiries. Every table is scoped by organization_id for multi-tenancy, and the dashboard's analytics are SQL views over that data. The backend owns the Aurora connection pool; the frontend talks only to the backend's REST API.

Challenges we ran into

The biggest was scoping. Our first architecture draft tried to rebuild the entire chatbot on managed AWS services — including OpenSearch Serverless, which alone starts around $700/month and would have meant throwing away a working RAG pipeline. We made the harder, better call: keep the brain we have, and use Aurora only for the relational data it's genuinely best at. Designing a clean multi-tenant schema with real foreign-key integrity — and analytics views that the dashboard could read directly.

Accomplishments that we're proud of

A real product, not a demo: the student experience is live and grounded, with an explicit anti-hallucination contract. A genuine B2B turn: the same system serves students and gives institutions a real-time view of student needs, with a clean closed loop (admin posts → students see). A deliberate, multi-tenant data model on Aurora that makes "one deployment, many universities" real, not aspirational. Cost discipline: prompt caching plus the right database for the job, so the whole thing runs cheaply.

What we learned

That the impressive-looking architecture isn't always the right one. Choosing Aurora over a flashier NoSQL or search-heavy stack — because our value is in relational analytics and our teammate already knew SQL — let us ship something solid in the time we had. We also learned how much leverage a multi-tenant schema gives you: one organization_id column turned a single-school tool into a licensable platform.

What's next for Student Companion AI Platform

Deploy the backend container to AWS App Runner for a fully AWS-native runtime. Onboard a second institution to prove the multi-tenant model end-to-end. Add semantic search over the structured knowledge base (pgvector on Aurora) and an admin view of cache-hit rate and cost over time. Expand the opportunities engine into a continent-wide feed to help close the job-opportunity gap for young Africans.

Built With

Share this project:

Updates