Skip to content

README.md

Latest commit

 

History

History
137 lines (95 loc) · 4.77 KB

README.md

File metadata and controls

137 lines (95 loc) · 4.77 KB

🤖 UBi – Agentic AI-Chatbot of UB Mannheim

UBi is an agentic AI-powered assistant for the Universitätsbibliothek Mannheim, built with Chainlit and LangChain. It combines large language models (LLMs) with data from the library website to deliver context-aware answers.

🚀 Features

  • 🧭 Agentic Router – Dynamically detects language, augments user queries, and intelligently routes them to the most suitable tool
  • 🧠 Semantic Augmentation – Enhances questions with context to optimize semantic search and retrieval
  • 🔌 Tool selector – Routes queries to one of three specialized tools:
    • 📖 RAG Pipeline – Retrieval-Augmented Generation using OpenAI embeddings, OpenAI inference, and OpenAI Cloud-based vectorstore
    • 📰 Library News Fetcher – Retrieves the latest updates directly from the UB Mannheim blog
    • 🪑 Real-time Seat Availability – Displays real-time information on study space availability at the library
  • 🌍 Multilingual Support – Detects and processes user input in multiple languages
  • 📝 Feedback Collection – Stores user questions, answers, and satisfaction ratings for continuous improvement
  • 📄 Terms of Use Popup – Ensures users accept terms before interaction
  • 🔐 Optional Login System – Supports password-protected access for restricted deployments

🛠 Tech Stack

Component Technology
Frontend UI Chainlit
Backend Logic Python + LangChain
LLMs OpenAI
Embeddings OpenAI
Vector Database OpenAI
Deployment Docker + Docker Compose

⚙️ Native installation

1. Clone the repository

git clone https://github.com/UB-Mannheim/UBi.git
cd UBi/code

2. Move .env.template to .env and add your OpenAI API key

OPENAI_API_KEY=sk-...

3. Install dependencies

pip install -r requirements.txt

4. RAG Pipeline Options

You can choose between two ways of running the app:

  1. Running the RAG pipeline locally
    • This option will embed all documents locally using the OpenAI embedding model text-embedding-ada-002 and create a chromadb vectorstore.
  2. Running the RAG pipeline with an OpenAI vectorstore

4.1 Running the app with local RAG pipeline

No OpenAI API key? Use Ollama instead. If OPENAI_API_KEY is left empty (or set to the placeholder sk-), UBi automatically falls back to Ollama for both the LLM and embeddings. Make sure Ollama is running locally and the required models are pulled:

ollama pull llama3.2
ollama pull nomic-embed-text

You can customise the models and the Ollama URL via the OLLAMA_ROUTER_MODEL, OLLAMA_CHAT_MODEL, OLLAMA_EMBEDDING_MODEL, and OLLAMA_BASE_URL variables in .env.

Start the app
chainlit run app.py
Open the app

Open http://localhost:8000 in a browser.

4.2 Running the app with an OpenAI vectorstore

Set USE_OPENAI_VECTORSTORE='True' in .env
OPENAI_API_KEY=sk-...
USE_OPENAI_VECTORSTORE='True'
Start the app
chainlit run app.py
Open the app

Open http://localhost:8000 in a browser.

🐳 Docker Deployment

1. Move .env.template to .env and add your OpenAI API key

OPENAI_API_KEY=sk-...
USE_OPENAI_VECTORSTORE='True' # Optional (for use with OpenAI vectorstore)
  • Optionally, set the exposed TCP port using the environment variable PORT (default: 8000).

2. Build and run

docker-compose up --build

3. Use

Open http://localhost:8000 in a browser.

💬 Feedback Logging

All chats and feedback are stored in the database data/feedback.db:

Field Description
session_id Random session ID
question User input
augmented_question Augmented user input
answer LLM-generated response
timestamp UTC datetime
feedback Score + optional comment

You can view or export this data for improving the bot.

License

This work is licensed under the MIT license (code) and Creative Commons Attribution 4.0 International license (for everything else). You are free to share and adapt the material for any purpose, even commercially, as long as you provide attribution.