Docu-Chat

The simple answer:

Docu-Chat is an AI-powered query engine that helps you extract insights from PDF documents. To use it:

1. Select a PDF file

2. Provide your OpenAI API key (see What is an "OpenAI API Key" and where do I get one? below)

3. Type a question related to the content of your document

The app will analyze the document and respond with an answer. It's like having a conversation with your document.

The technical answer:

Docu-Chat leverages several technologies to provide an interactive chat interface for querying PDF documents.

• PDF Handling: The user uploads a PDF file through a Panel file input widget. The app processes it using the PyPDFLoader class to load the document.

• LangChain Integration: LangChain is used to handle the retrieval of answers from the documents. The loaded PDF is split into smaller chunks using CharacterTextSplitter, and the chunks are then embedded using OpenAI embeddings. 

• Chroma Vector Store: Chroma is utilized to create a seamless vector store from the embedded documents, which is then used as a retriever for similarity searches.

• Query Processing: The user's query is processed, and any empty lines are removed. If there is a chat history, it's appended to the query to provide context. The RetrievalQA class from LangChain is used to perform a similarity search on the documents, and the retrieved answer is returned to the user. This app currently uses LangChain's "stuff"  chain to perform the query.

• OpenAI Integration: An OpenAI API key is required for the embeddings. The key is entered by the user and set as an environment variable to authenticate with OpenAI services.

• In-Memory Conversation History: The app maintains a chat history in memory, allowing the context of previous interactions to be considered in subsequent queries.

• Hugging Face Hosting: The entire app is hosted on Hugging Face, providing seamless integration between LLMs and open source access to the code. See Why did you build this? for the link to source code.

At the moment Docu-Chat is designed to analyze PDF files only. For best results, upload PDFs with machine readable text. Scanned documents and PDFs without machine readable characters will result in lower quality results. There's no limit to the size or length of file you upload, but very large files will require more time to process and more OpenAI tokens (see What is an "OpenAI API Key" and where do I get one?).

Please note that although this application has been designed with user privacy and data security in mind, it is not recommended to upload sensitive documents. See the full app disclaimer at the bottom of this page.

Good question. Here's how do I get your API Key:

This is your unique key (password) which authorizes your use of OpenAI (the clever engine that powers Docu-Chat). You can get your unique key by doing the following:

1. Create an OpenAI account‍, here ⧉ (if you haven't already got one)

2. Navigate to the API section, here ⧉

3. Generate a new API Key and save it somewhere safe

4. Set up billing, and set usage limits (optional but recommended)

Once you have a key, you'll need to enter it into this application before you can use it – that way, OpenAI knows who's making the request!

But what is an API?

An API (Application Programming Interface) is a set of rules that allow different software applications (like Docu-Chat and OpenAI) to communicate with each other. The API defines how requests and responses (like the questions you ask) should be formatted, and what data (like the documents you upload) can be accessed.

And what is an API Key?

An API Key is basically your unique password that allows you to access the specific services provided by the API. The key helps the API provider (in this case OpenAI) to identify you, track your usage, and control access to the data and functions. Essentially, the API Key ensures that only authorized users can interact with OpenAI.

Yes and no. Let me explain:

In simple terms, the app itself is totally free to use. I don't charge anything for the app and don't make any money from it (in fact, I pay to keep it running)! But this application uses OpenAI to process your requests. OpenAI charge a very small amount of money for this, and you'll need to provide an OpenAI API Key before using this application. OpenAI will charge you approximately $0.03 for every 15,000 words you upload, and $0.04 for every 15,000 words of analysis generated. Since the average length of book is 90,000 words, it's about 18 cent to analyze a whole book (...pretty cheap).

The more technical explanation:

Docu-Chat uses OpenAI's large language model (GPT-3.5-Turbo) to process your requests via API. OpenAI charge users to make requests via API, and that's why you need to provide your unique API Key. When you make an API request, OpenAI calculate a fee based on the number of tokens used. Tokens can be thought of as a number of characters. Before the API processes your request, your input is broken down into tokens. Here are some helpful rules of thumb for understanding tokens in terms of lengths: 1 token = ~4 characters in English. 100 tokens = ~75 words. 2 sentence = ~30 tokens. 1 paragraph = ~100 tokens. 1,500 words = ~2048 tokens.

This application currently uses OpenAI's GPT-3.5-Turbo model, which costs $0.0015 / 1K tokens (input) and $0.002 / 1K tokens (output). So if you upload 1,500 words (~2,048 tokens) you'll be charged less than 1/3 of a cent ($0.003). You can read more about OpenAI's model pricing here ⧉ and calculate the number of tokens in your text here ⧉. You can maintain control of your spending by reviewing your usage and setting spending limits via your OpenAI account. 

There's no strict limit on the number of questions you can ask. You can engage in an ongoing conversation with the AI to extract detailed insights from your uploaded document. To keep the query engine efficient, I have limited the "working memory" so if you ask multiple questions, the engine will "forget" the older questions you've asked (see How does Docu-Chat work?)

DocuChat uses advanced AI algorithms to analyze the content of your uploaded PDF and provide answers to your questions. While it's designed to be highly accurate, it's always recommended that you cross-reference critical information with the original source. The AI's interpretation is based on the content within the document, but it can make mistakes.

Great question.

I built Docu-Chat to empower users with instant access to insights hidden within long and/or complex documents. In today's information-rich world, finding specific answers within extensive documents can be time-consuming. Docu-Chat improves efficiency and streamlines information extraction.

Other apps might already do this, but they charge premium subscription fees and put restrictive limits on a users. For example, other application cap your number of uploads and number of queries. By making an open source free-to-use version, you don't pay any unnecessary costs, you aren't locked into a subscription, and you're very welcome to clone, remix, re-purpose, or upgrade this app yourself via Hugging Face ⧉

That's an easy one.

Other apps charge subscription fees and restrict your usage (for example, they limit the number of uploads, your file sizes, and the number of queries you can make). They also lock you into subscriptions, so you pay for a whole month or year of usage, even if you only analyze one file. A number of these apps are also using older AI models, resulting in lower quality answers than Docu-Chat can provide.

This application free to use and open source. You just pay for your OpenAI usage (which you're in control of). No subscriptions, no usage fees, no service costs, and no user limits. And because it's open source, you're welcome to clone, remix, re-purpose, or upgrade this app however you wish! Full code is available via Hugging Face ⧉

Feel free to GET IN TOUCH if you're interested in learning more about this application, or using generative AI.