Praxa 🎭

A RAG-powered chatbot for answering questions about West End and Broadway theatre. Praxa combines a curated PDF knowledge base with a retrieval-augmented generation pipeline to provide accurate, source-backed answers about shows, productions, venues, and the people behind them.

---

Features

• Conversational chat interface built with Streamlit
• Retrieval-Augmented Generation (RAG) — answers are grounded in your own documents, not just LLM training data
• Source citations with every answer (filename + page number)
• Local embeddings via all-MiniLM-L6 — no API call needed for indexing
• Powered by gemma-3-27b via OpenRouter

---

Tech Stack

Layer	Tool
UI	Streamlit
LLM Orchestration	LangChain
LLM	gemma-3-27b via OpenRouter
Embedding Model	all-MiniLM-L6-v2 (sentence-transformers)
Vector Database	Chroma
Document Loader	LangChain PyPDFLoader
Text Splitting	RecursiveCharacterTextSplitter

---

Project Structure

Praxa/
├── praxa_client.py     # Streamlit UI — chat interface
├── praxa_rag.py        # RAG chain — retrieval, prompt, LLM, sources
├── context.py          # Chroma vector store setup and PDF indexing
├── model.py            # OpenRouter LLM initialisation
├── data/               # PDF documents (your knowledge base)
├── chroma_db/          # Persisted Chroma vector index (auto-generated)
└── requirements.txt

---

Setup & Installation

1. Clone the repository

git clone https://github.com/your-username/praxa.git
cd praxa

2. Create and activate a virtual environment

python -m venv venv

# Windows
venv\Scripts\activate

# macOS / Linux
source venv/bin/activate

3. Install dependencies

pip install -r requirements.txt

4. Set your OpenRouter API key

Create a .env file in the project root:

OPENROUTER_API_KEY=your_api_key_here

You can get a free API key at openrouter.ai.

5. Build the vector index

Run the following command to set up Praxa's knowledge base. This will automatically download the required PDFs and build the Chroma vector index:

python context.py

This only needs to be done once. The PDFs will be saved to context_data/ and the vector index to chromadb/.

---

Running the App

streamlit run praxa_client.py

Then open http://localhost:8501 in your browser.

---

How the RAG Pipeline Works

User question
      │
      ▼
 Embed question           ← all-MiniLM-L6-v2
      │
      ▼
 Similarity search        ← Chroma finds top-k most relevant chunks
      │
      ▼
 Build prompt             ← LangChain formats question + retrieved context
      │
      ▼
 LLM generates answer     ← gemma-3-27b via OpenRouter
      │
      ▼
 Return answer + sources  ← filename and page number for each chunk used

At indexing time, each PDF is loaded, split into overlapping chunks using RecursiveCharacterTextSplitter, and embedded using all-MiniLM-L6-v2. The resulting vectors are stored in a local Chroma database alongside the original text and metadata.

At query time, the user's question is embedded using the same model, and Chroma performs a cosine similarity search to retrieve the most relevant chunks. These are passed to gemma-3-27b as context alongside the question, and the model generates a grounded answer. The source documents are returned alongside the answer so the user can verify the information.

---

Requirements

streamlit
langchain
langchain-community
langchain-chroma
langchain-openai
sentence-transformers
chromadb
pypdf
python-dotenv

---