Aaltoes AI Search

Aaltoes AI Search is a Retrieval-Augmented Generation (RAG) application designed to facilitate intelligent search and question-answering over documents related to Aaltoes (Aalto Entrepreneurship Society). The application leverages Google Drive as a document source, ChromaDB as a vector store, and integrates with state-of-the-art language models like OpenAI's GPT and Google's Gemini.

Features

• Document Retrieval: Fetch documents from Google Drive based on year and file type.
• Question-Answering: Generate answers to user queries using retrieved documents.
• Embedding Models: Support for OpenAI and Google embedding models.
• Language Models: Integration with OpenAI GPT models and Google's Gemini models.
• Document Parsing: Extract and process content from various file types, using unstructured library.
• Translation: Translate Finnish documents to English for better accessibility.
• Gradio Interface: User-friendly web interface for querying and exploring results.
• FastAPI Integration: Expose indexing and retrieval functionalities via a FastAPI-based REST API.
• Concurrency and Parallelism: Leverage asynchronous tasks and process pools to speed up document indexing and retrieval.
• Tracking Indexed Documents : Track indexed files and enable checkpoint recovery.

TODO

• Agentic coding for spreadsheets

Prerequisites

• Python 3.12 or higher.
• uv as the project manager.
• Google Cloud credentials for accessing Google Drive.
• API keys for OpenAI and Google Gemini.

Installation

1. Clone the repository:

   git clone <repository-url>
   cd <repository-folder>

2. Set up your environment:

   • Create a .env file in the project root.
   • Add the following variables:

     GEMINI_API_KEY=<your-google-gemini-api-key>
     OPENAI_API_KEY=<your-openai-api-key>
     CHROMADB_PERSISTED_PATH=./chroma_db

3. uv will create the venv and install dependencies for you when you run the app.

Quick Start

Using the CLI

1. Index Documents: To index documents from Google Drive from previous, run:

   uv run main.py --mode index --resume_checkpoint True

   use --help to see other options

2. Retrieve Documents: To retrieve documents from Google Drive, run:

   uv run main.py --mode retrieve --query <your query>

   use --help to see other options

Using Gradio

Search and Retrieve: Use the Gradio interface to input your query, select parameters like year, model, and embedding function, and retrieve answers along with document references. To launch the Gradio interface, run: bash uv run gr_app.py

Using the FastAPI Endpoints

1. Start the FastAPI Server: Run the FastAPI app in dev mode:

   fastapi dev fapi_app.py

   for production mode:

   fastapi run fapi_app.py

2. Index Documents: Send a POST request to the /index endpoint with the required parameters. Example request body:

   {
       "mode": "index",
       "model": "gpt-4o",
       "emb_func": "openai",
       "indx_years": [2020, 2021]
   }

3. Retrieve Documents: Send a POST request to the /retrieve endpoint with the required parameters. Example request body:

   {
       "mode": "retrieve",
       "model": "gpt-4o",
       "emb_func": "openai",
       "query": "What was the purpose of Aaltoes?",
       "top_k": 5,
       "retr_year": "2021"
   }

   Example response:

   {
       "response": "Aaltoes was focused on fostering entrepreneurship...",
       "reference_files": "- Document 1: \"Event Summary 2021\" from year 2021\n- Document 2: \"Partnerships 2021\" from year 2021"
   }

4. Check API Status: Access the root endpoint to verify the API is running:

   curl http://127.0.0.1:8000/

Configuration

The application uses a APIConfig class (based on Pydantic's BaseModel) to validate and manage configuration parameters. Key parameters include:

• mode: Operation mode (index or retrieve).
• model: Language model to use (e.g., gpt-4o, gemini-2.0).
• emb_func: Embedding function (openai or google).
• indx_years: List of years to index or "Full" for all years.
• query: Query string for retrieval.
• top_k: Number of top results to retrieve.
• retr_year: Year to filter retrieval results or "Full" for all years.
• file_type: File type to filter (e.g., PDF, MSWords, or "Full" for all types).

Notes

• The FastAPI app provides a programmatic interface for indexing and retrieval, making it easier to integrate with other systems.
• The CLI remains available for quick local testing and usage.
• Ensure that the .env file is properly configured with valid API keys and paths before running the application.