This project provides a system for ingesting and querying PDF documents using Retrieval-Augmented Generation (RAG) techniques. The system leverages FastAPI for the backend, integrating with Inngest for event-driven workflows and utilizing Google's Gemini API for content generation and embedding. It uses Qdrant for efficient vector-based search, enabling scalable PDF ingestion and AI-driven querying of document contents.

• PDF Ingestion: Extracts and chunks text from PDFs and stores the embeddings in a vector database (Qdrant).
• AI-Powered Queries: Allows users to query the ingested PDFs using a question, retrieving relevant document chunks using semantic search.
• Event-Driven: Utilizes Inngest for scalable and asynchronous event handling.
• Scalable Search: Leverages vector embeddings for high-quality, fast search results.

• Python 3.8+
• Dependencies managed with uv (instead of pip)
• Environment variables

1. Clone the repository:

 git clone https://github.com/your-repo.git
 cd your-repo

2. Install Dependencies The project uses uv for package management, so use the following command to install dependencies:

uv install

3. Create a .env File Configure the following environment variables:

• GEMINI_API_KEY=<Your-Gemini-API-Key>
• QDRANT_API_KEY=<Your-Qdrant-API-Key>
• QDRANT_URL=<Your-Qdrant-URL>

4. Running the Application To run the application locally, use the following command:

uv run uvicorn main:app --reload

The backend will be available at http://localhost:8000.

Endpoint: PATCH /ingest-pdf

Description: Ingests a PDF document, extracts text, chunks it, and stores the embeddings in Qdrant for future querying.

Request Body:

{
  "pdf_path": "<path-to-pdf>",
  "source_id": "<optional-source-id>",
  "collection": "<collection-name>"
}

Response: A summary of the ingestion process, including the number of chunks ingested.

Endpoint: POST /query-pdf-ai

Description: Queries the ingested PDFs using a question, returning relevant document chunks and sources.

Request Body:

{
  "question": "<user-question>",
  "top_k": <number-of-results>,
  "collection": "<collection-name>"
}

Response:

{
  "answer": "<AI-generated-answer>",
  "sources": ["source1", "source2"],
  "num_contexts": <number-of-contexts>
}

When a PDF is uploaded, an ingest_pdf event is triggered. The document is:

• Loaded
• Chunked
• Embedded
• Stored in the Qdrant vector database

tWhen a query is made, a query_pdf_ai event is triggered. The system: e1. Performs a vector search to find the most relevant document chunks e2. Uses Gemini AI to generate a concise answer based on the retrieved context.