A learning platform where a mentor builds a course out of groups, modules and tasks, and learners get their answers and code checked the moment they submit.





Study groups, clubs and small courses usually live in a chat plus a shared spreadsheet. The mentor posts a task, collects answers by hand, reruns every piece of code on their own machine and keeps scores in a table nobody else can see. Oriole puts that loop in one place: the mentor creates a space, invites people with a link, publishes tasks in order, and the platform grades what it can on its own so the mentor only reviews what needs a human.
%%{init: {"theme":"base","themeVariables":{"primaryColor":"#ffffff","primaryTextColor":"#141317","primaryBorderColor":"#7a3ec9","secondaryColor":"#efedf3","tertiaryColor":"#f7f6f9","lineColor":"#67646d","textColor":"#141317","noteBkgColor":"#efedf3","noteTextColor":"#2f2d35","actorBkg":"#ffffff","actorBorder":"#7a3ec9","fontFamily":"Montserrat"}}}%%
flowchart LR
M[Mentor] -->|creates| S[Space]
S --> G[Group]
G --> Mod[Module]
Mod --> T[Task]
L[Learner] -->|joins by link| G
L -->|submits| Sol[Solution]
Sol --> API[FastAPI]
API -->|answer tasks| C1[Instant check]
API -->|code tasks| Q[RabbitMQ]
Q --> W[Celery worker]
W -->|one container per test| R[Python or Node runner]
R --> V[Verdict]
C1 --> V
- Spaces, groups and modules. A space is the top level: a base space, a group or a course. Groups hold members with roles; modules order the tasks inside a group. People join by invite link or by a request the owner approves.
- Seven task types. Free answer, string match, multiple choice, file upload, code, AI-assisted and combined tasks share one model, so a module can mix them.
- Code checks in isolation. A code solution goes to a Celery worker through RabbitMQ. The worker starts a fresh container per test with the task's memory and time limits, compares the output and stores one verdict: accepted, wrong answer, runtime error, time limit or memory limit exceeded.
- Accounts. Sign-up with email confirmation, password reset, email change and login through Telegram. Tokens are RS256 JWTs with refresh.
- Group chat. A WebSocket channel per group, with history kept in PostgreSQL. The chat client
lives in
src/chat/. - Limits and caching. Redis backs response caching and a per-client rate limiter.
- Observability. The API exposes Prometheus metrics; Promtail ships container logs to Loki; Grafana comes up with a provisioned dashboard.
You need Docker with Compose and make.
git clone https://github.com/GoatWhistle/oriole.git
cd oriole
cp src/backend/.env.example src/backend/.env
cp src/backend/.env.app_config.example src/backend/.env.app_config
openssl genrsa -out src/backend/jwt-private.pem 2048
openssl rsa -in src/backend/jwt-private.pem -pubout -out src/backend/jwt-public.pem
make upFill the ports, database credentials and SMTP account in both .env files before make up. The web
app opens at http://localhost, the API at http://localhost:<BACKEND_PORT>. Code tasks need the
worker to reach the host Docker socket. Details, service ports and troubleshooting are in
docs/running.md.
| Command | What it does |
|---|---|
make up |
Builds images and starts the stack |
make down |
Stops the stack |
make re |
Rebuilds and restarts |
make test |
Runs backend tests from src/tests |
All routes live under /api. With IS_DEV=true the backend serves Swagger UI at
http://localhost:<BACKEND_PORT>/docs.
| Prefix | Purpose |
|---|---|
/api/auth, /api/auth-via-telegram |
Login, refresh, Telegram login |
/api/users, /api/verify |
Registration, profile, email confirmation and password reset |
/api/spaces, /api/groups |
Spaces, groups, invites, join requests, roles |
/api/modules, /api/tasks |
Modules, tasks of every type, solutions |
/api/chat, /api/websocket |
Group chat history and live channel |
/metrics |
Prometheus metrics |
src/
backend/ FastAPI app: features/, core/ (config, Celery, Redis, logging), alembic/
frontend/ React 19 web app with Ant Design
chat/ React client for the group chat
tests/ unit and integration tests
docker/ backend and frontend Dockerfiles, nginx config
monitoring/ Prometheus, Loki, Promtail and Grafana provisioning
docker-compose.yaml
- The code runner supports Python and JavaScript only, and it needs access to the host Docker socket, which gives the worker full control of the Docker daemon. Run it on a machine you trust.
- AI-assisted and combined task types exist in the data model; automatic grading covers string match, multiple choice and code.
- The chat client is a separate development app and points at
127.0.0.1:8000. - The repository ships no production setup: no TLS, no deployment config, development defaults for RabbitMQ and Grafana credentials.

Team GoatWhistle, 2025. Licensed under MIT.












