Case study · 02 / 10

UWERA AI Tutor

A conversational tutor for Rwandan students, grounded in the curriculum with retrieval.

Organisation: Rwanda Coding Academy
Role: AI engineer
When: Mar 2026 to now
Live: uwera.rw (opens in a new tab)

UWERA website: AI powered learning built for every student, with two students working on a laptop — UWERA AI Tutor, public site

Context

Students need answers they can trust. A general chatbot invents things. UWERA has to stay inside the curriculum, remember the conversation, and run on open source models.

UWERA AI Tutor on a phone — The same site on a 390 px phone screen.

What I built

01Retrieval augmented generation pipelines over course material, with open source LLMs.
02A multi turn context engine that keeps long tutoring sessions coherent.
03Multi provider model APIs, file storage and chat history persistence behind a Next.js interface.

System view

Architecture of UWERA AI Tutor, drawn as a diagram

Outcome

45% fewer hallucinations from the language model.
30% higher weekly engagement.

Tools

Python
FastAPI
Next.js
LangChain
Hugging Face
Milvus
PostgreSQL