# Chat with documents using LLM

This is a Bellingcat [hackathon](https://www.bellingcat.com/april-hackathon-announcement-mar-2023/) submission

## Team Members

[Radu Ciocan](https://www.linkedin.com/in/ciocan/) - code

[Ana State](https://www.linkedin.com/in/anastate/) - design

## Tool Description

A chat interface that allows users to upload PDF/DOCX documents and and chat with them using ChatGPT as the large language model.

https://user-images.githubusercontent.com/4984377/233852126-6828f7b6-cfc3-485c-a893-c2ce9e6b345f.mp4

## Installation

1. Make sure you ave installed Node v18 or later
2. Download the tool's repository using the command:

```bash
git clone git@github.com:ciocan/langchain-chat-with-documents.git
```

3. Move to the tool's directory and install the tool

```bash
cd langchain-chat-with-documents
npm install
```

4. Copy the .env.example into .env file and add the following variables:

```bash
WEAVIATE_HOST= # do not use https:// just the domain like bellingcat-xxx.weaviate.network
WEAVIATE_API_KEY=

# cloudflare r2
CLOUDFLARE_ACCOUNT_ID=
CLOUDFLARE_SECRET_KEY=
CLOUDFLARE_SECRET_ACCESS_KEY=

# open ai key
OPENAI_API_KEY=
```

Weaviate is an open source vector database where the documents are vectorized and indexed. You can install it locally or use their [free cloud](https://console.weaviate.cloud/).

Cloudflare R2 is a object storage solution compatible with AWS S3. They have a free tier of 10gb. [More info](https://www.cloudflare.com/products/r2/)

Signup for an OpenAI API key [here](https://platform.openai.com/)

1. Start the tool

```bash
npm run dev
```

## Tech Stack

This is a [T3 Stack](https://create.t3.gg/) project bootstrapped with `create-t3-app`.

- [Next.js](https://nextjs.org)
- [Tailwind CSS](https://tailwindcss.com)
- [tRPC](https://trpc.io)

Additional libraries:

- [Zustand](https://github.com/pmndrs/zustand) - Used for state management
- [Mantine UI](https://mantine.dev) - Used for awesome UI components
- [LangChain](https://js.langchain.com/docs/) - Used for interacting with the LLM model (OpenAI)

## High Level Architecture

![architecture](https://user-images.githubusercontent.com/4984377/233864756-ae1fc84b-2c64-490f-892f-022bd1c1a8b7.png)