Official repository of master's thesis "Implementing a Retrieval Augmented Generation System" (2025) by David Slavik, student of University of Zagreb Faculty of Informatics.
Goal of the project is to build a RAG system which will be useful for providing answers to questions regarding a codebase of a small to medium scale project.
Users can upload their codebase as a zip file or provide the GitHub repo link. After that, all the relevant files in the provided codebase are downloaded on backend and the relevant ones such as code, text, images are indexed in Postgres database (thanks to PgVector extension).
Once the indexing of files is finished, users can start asking questions in a chat interface. LLM which is used is qwen/qwen3-32b by Alibaba Cloud because of its logical reasoning and coding capabilities. Because the documents are indexed in PgVector document store, agent will understand the context of provided codebase and will take it into account when answering queries.
Conversation history is stored and users can bookmark or delete their conversations.
React - JS library for creating user interfaces
Vite - JS bundler
TailwindCSS - framework which allows CSS styling directly within HTML by providing utility classes
shadcdn/ui - free and open source accessible and customizable UI components
Postgres - RSQL database which supports storing vector embeddings using PgVector extension
FastAPI - framework for creating API in python
Haystack - python framework for building AI pipelines and applications
Docker - for automated deployment of applications in containers
Groq - low-cost, high performance inference platform which provides API to various LLMs
- Docker
- Docker Compose
- Git (if cloning the repository)
You do not need to install Python, Node, or other language runtimes manually - the services are containerized
Before building, it is necessary to create a .env file in root directory of the project which will contain environment variables and their values. Add your values without <> signs.
GROQ_API_KEY=<your_groq_api_key>
VOYAGE_API_KEY=<your_voyage_api_key>
VOYAGE_MODEL=<voyage-code-2 OR voyage-code-3>
POSTGRES_USER=<your_db_user>
POSTGRES_PASSWORD=<your_db_password>
POSTGRES_DB=<your_db_name>
DATABASE_DSN=postgresql://<user>:<password>@<host>:<port>/<dbname>
JWT_SECRET=<replace_with_a_random_secret>
JWT_ALGORITHM=HS256
ACCESS_TOKEN_EXPIRE_MINUTES=1440
UPLOAD_DIR=uploads
DATA_DIR=data/reposTo build the application, simply run the following command:
docker-compose up --build.
This command will install necessary tools and build frontend & backend images from instruction files Dockerfile.frontend, Dockerfile.backend.
Backend will be available on port 3001 and frontend on port 5173.
To start using the app, just visit http://localhost:5173.
This project is licensed under the MIT License. See the LICENSE file for details.