🚀 Chatten

RAG with sources, built with Dash, FastAPI, and the Databricks platform.

---

💫 Showcase

Here is an example of the app in action:

<div style="text-align: center;"> <img src="assets/showcase.gif" alt="chatten" style="max-width: 800px; height: auto;"> </div>

🛠 Developer Setup

To install the project, ensure you have the following dependencies:

• 📦 uv: for managing the project
• 🚀 Databricks CLI: for deploying the app
• 🌐 Node.js: for building the UI

📥 Installation Steps

1. Clone the repo:

   git clone <repo-url>
   

2. Run sync:

   uv sync --all-packages
   

3. Configure environment variables in the .env file:

   # Name of your Databricks profile
   CHATTEN_PROFILE=...
   
   # Optionally, you can include any bundle or Chatten variables:
   CHATTEN_CATALOG=...
   
   # Optionally, by check default values in databricks.yml
   BUNDLE_VAR_vsi_endpoint=...
   

---

🏗 Development

1. Start the UI watcher in one terminal:

   cd packages/chatten_ui && npm run watch
   

2. Run the server in another terminal:

   uvicorn chatten_app.app:app --reload
   

---

🚀 Deployment

1. Authenticate with Databricks:

   databricks auth login -p <profile-name>
   

2. Deploy the app:

   # See Makefile for additional variables
   make deploy profile=fe-az-ws catalog=<catalog-name>
   

3. Run the RAG workflow:

   # See Makefile for additional variables
   make run-rag profile=fe-az-ws catalog=<catalog-name>
   

4. Grant app principal access to the Volume.

5. Run the app:

   make run-app profile=fe-az-ws catalog=<catalog-name>
   

6. Open the app from Databricks Workspace 🎉

---

🤖 Agent Serving Endpoint Response Parsing

Check the api_app source code for details on how the agent serving endpoint response is parsed.

Specifically, the /chat API endpoint handles:

• Parsing responses
• Sending messages to the chat interface

---

🏛 App Implementation Details

The implementation consists of a FastAPI backend with two sub-apps:

1. Dash app (/ route): Uses a custom component to render the chat UI.
2. FastAPI app (/api route): Provides API endpoints for the chat.

The chat is implemented as a custom Dash component, which is a React-based UI element that communicates with the FastAPI backend.

The /api app interacts with the Databricks Serving Endpoint to handle chat requests and responses. Another route is responsible for serving PDF files from Databricks Volume.

---

📂 Code Structure

📦 chatten  # Main package (FastAPI + Dash app)
 ┣ 📂 packages/chatten_ui  # Dash UI & custom chat component
 ┣ 📂 packages/chatten_rag  # RAG workflow implementation
 ┣ 📂 packages/chatten_app  # FastAPI & Dash apps
 ┣ 📂 src/chatten  # Common config & utilities
 ┣ 📂 src/chatten/app.py  # Main entry point

---

🏗 Technologies Used

🔥 Core Platform

• Databricks
  • Apps - App serving
  • Asset Bundles - Deployment
  • Mosaic AI Model Serving - Model serving
  • Mosaic AI Vector Search - Vector search

🏗 Frameworks & Libraries

• 🖥 Dash - UI framework
• ⚡ FastAPI - Backend API
• 🎨 Tailwind CSS - Styling
• 📄 react-pdf - PDF rendering
• 🔍 Fuse.js - Client-side fuzzy search
• 📜 pypdf - Server-side PDF text extraction
• ⚡ RapidFuzz - Fuzzy string matching
• 📦 uv - Dependency management