This project has been created as part of the 42 curriculum by pjarnac, dernst, dderny, tjooris.

KeepGrouped

Description

KeepGrouped is a web platform to organize events and keep their participants together in one place. Each event gets its own chat channel, shared files and an AI assistant that answers questions from the event documents.

Key features:

• Event creation, registration and calendar
• Real-time chat (channels, categories, direct messages, read state)
• Friends, blocking and user profiles with avatars
• File upload and management (S3-compatible storage)
• AI chatbot with RAG over uploaded documents and events.
• Roles, permissions and an admin panel
• Interface in English, French and Chinese

Instructions

Prerequisites:

• Git
• Docker Engine 29.3 or newer
• Docker Compose v5.1 or newer
• OpenSSL, to generate the secrets
• A machine with at least 4 GB of RAM, or a GPU with 4 GB of VRAM, for the LLM

1. Clone this repository.

2. Create the environment file and fill every value in it: port, server name, database password, Garage access/secret/RPC keys, admin login, email and password, JWT key. The comments in .env.example give the command to generate each secret.

   cp .env.example .env
   

3. Start everything:

   docker compose up -d
   

4. Open https://localhost:4242 (or the PORT set in .env) and log in with the default admin account. The TLS certificate is self-signed, so the browser shows a warning that you have to accept.

The source code of each service lives in its own repository: frontend, backend, backend-ia, keepgrouped-llama.cpp.

Team Information

Login	Role(s)	Responsibilities
pjarnac	Product Owner & Developper	Events, file storage, AI integration in the backend, infrastructure and CI, invitations
dernst	Project Manager & Developper	Authentication, users, roles, admin panel, multi-language
dderny	Technical Lead & Developper	Chat, channels, friends and relationships
tjooris	Developper	AI service (RAG pipeline, chatbot)

Project Management

• Organization: Work dispatched by big features. PR Reviews
• Tools: Forgejo (Git hosting, CI, container registry, issues).
• Communication: Discord on a daily basis.

Technical Stack

Layer	Technologies
Frontend	React 19, TypeScript, React Router, Vite, Tailwind CSS, TanStack Query, Zustand, i18next, SignalR
Backend	ASP.NET Core (.NET 11), Entity Framework Core, SignalR, JWT auth
AI service	Rust, Axum, Rig, FastEmbed, ChromaDB
LLM	llama.cpp server running Gemma 4 E2B (quantized)
Database	PostgreSQL
Storage	Garage (S3-compatible)

Justification:

• PostgreSQL: relational data (users, events, channels, roles) with many relations and constraints; first-class EF Core support.
• ASP.NET Core + SignalR: typed, fast backend with built-in real-time transport.
• Rust for the AI service: low memory footprint and good async support for document parsing and streaming responses.
• Local LLM with llama.cpp: no external API, data stays on the server.
• Garage: lightweight self-hosted S3 storage for files and avatars.

Database Schema

Main tables (PostgreSQL, managed by EF Core migrations in backend/Migrations):

Table	Key fields	Relations
Users	Id (string), UserName, Email, PasswordHash, Language (enum), IsOnline (bool)	Role (n-1), Avatar file (1-1)
Roles	Id, Name, Ranking, Permission (flags)	Users (1-n)
Events	Id, Name, Date, Size (int), Location, Tags (string[]), Description	Organizer (n-1), Channel (1-1), Users via Registrations (n-n), EventRoles, Files
EventRoles	Id, Name	Events (n-n)
Channels	Id, Type (enum), Name, Topic, Order	Category (n-1), Event (1-1), Members (1-n), RolesWhitelist (1-n)
ChannelCategories	Id, Name, Order	Channels (1-n), RolesWhitelist (1-n)
ChannelMembers, ChannelAcks	composite key (ChannelId, UserId)	Membership and last read message
ChannelRoles	ChannelId or CategoryId, RoleId	Role whitelist for channels and categories
Messages	Id, Content (max 8192), SentAt, MessageRefId	Channel (n-1), Sender (n-1), replied message
Relationships	MeId, UserId, Type (enum), Nickname, Since	Friends, pending requests and blocks
Files	Key, Name, Length, ContentType, ETag	Creator (n-1), Events (n-n)
Invitations, RefreshTokens	Id, ExpireAt, Usages	Platform invitations and sessions

Features List

Feature	Description	Member(s)
Authentication	Register, login, JWT + refresh tokens, sessions list	dernst
Users and profiles	Profile, avatar, language, password change	dernst
Admin panel	Manage users, roles and permissions	dernst
Invitations	Invite users to the platform	pjarnac
Events	Create, edit, register, calendar, event roles	pjarnac
Chat	Channels, categories, DMs, replies, read state, real time	dderny
Friends	Friend requests, blocking, online status	dderny
Files	Upload, rename, download, delete	pjarnac
Activity Status	Automatic update with browser activity, real time.	pjarnac
AI assistant	Chatbot answering from event documents (RAG)	tjooris, pjarnac
Translations	English, French, Chinese	dernst

Modules

Module	Type	Points	Implementation	Member(s)
Framework for frontend and backend	Major	2	React + ASP.NET Core	all
Real-time features	Major	2	SignalR hub for chat and activity	dderny, pjarnac
User interaction	Major	2	Chat, profiles, friends	dderny, dernst
Standard user management	Major	2	Auth, profiles, avatars	dernst, pjarnac
Advanced permissions system	Major	2	Roles, channel whitelists, admin panel	dernst
RAG system	Major	2	Rust service, FastEmbed + ChromaDB	tjooris
LLM system interface	Major	2	Streaming chatbot on llama.cpp	tjooris, pjarnac
ORM	Minor	1	Entity Framework Core	all
File upload and management	Minor	1	Garage S3 storage	pjarnac
Multiple languages	Minor	1	i18next, 3 languages	dernst
Custom-made design system	Minor	1	Buttons, various inputs, modals, badges, list, etc..	dernst, pjarnac, dderny
Support for additional browsers	Minor	1	Firefox, opera, chrome	all
Total		19

Individual Contributions

• pjarnac: events (backend and frontend), file storage with Garage, AI backend integration and RAG queue, llama.cpp image, Docker Compose, CI. Challenges: Creating a easily usable dev env with devcontainers was really hard, file management with the horrible Amazon S3 Documentation.
• dernst: authentication with JWT, users, roles, multi-languages, permissions, invitations, admin panel. Challenges: Did you know that handling permissions is definitely not easy? One tiny mistake and a simple member can become an admin. Add token management on top of that, and you have the reason behind my 4 months of nightmare.
• dderny: chat backend and UI, channels, friends and relationships. Challenges: Did you know that replicate Discord it's a huge work ? The more challenging was database structure and messages paginations.
• tjooris: Rust AI service, document parsing, embeddings, chatbot. Challenges: Learning Rust was tough, but ingesting files using multipart was by far the most challenging part.

Resources

• ASP.NET Core, EF Core, SignalR
• React, React Router, TanStack Query
• Rig, ChromaDB, llama.cpp
• PostgreSQL, Garage

AI usage: Used for late redesign of certains frontend components (Modals, Inputs, Buttons). Big scale refactor and file moves. Research. Bug finding. A part of this README.