ChatPDF with Google Gemini

A RAG to chat with multiple PDF files using Google Gemini LLM.

Requirements

The following components are required to run ChatPDF application.

Google Gemini API

You need to get an API key from Gemini. Google also provides free API keys.

MongoDB Atlas

We are storing the embeddings in MongoDB to perform Vector Search. This feature is available on cloud-based version of MongoDB, aka. MongoDB Atlas. You can create a free account on its website.

• Register for a free MongoDB Atlas Account
• Create a Cluster
• Get your connection string

Vector Search Index

Once a cluster is created on MongoDB Atlas, go ahead to create a database and a collection. A vector search index with the following format should be created on the created collection.

"fields": [
  {
      "type": "vector",
      "numDimensions": 768,
      "path": "embedding",
      "similarity":  "cosine"
  },
  {
      "type": "filter",
      "path": "file_name"
  }
  ]

• Create an Atlas Vector Search Index

Streamlit Secrets

A Streamlit secrets file should be created in .streamlit/secrets.toml with the following format, providing the details of the API key and MongoDB settings:

[Gemini]
GEMINI_API_KEY = "api_key_here"
LLM_MODEL = "gemini-2.0-flash"
EMBED_MODEL = "models/text-embedding-004"

[MongoDB]
URI = "mongodb+srv://<username>:<password>@<cluster_name>.qwar6.mongodb.net/"
DB = "pdf"
COLLECTION = "embeddings"

Run

In order to run this tool, you must have Streamlit installed on your machine/environment:

streamlit run app.py

Github Repo

This project is open-source, and it is available on Github at https://github.com/kavehbc/chat-pdf.

Screenshot

Developer(s)

Kaveh Bakhtiyari - Website | Medium | LinkedIn | GitHub

Contribution

Feel free to join the open-source community and contribute to this repository.