Search Docs using Open AI

This projects lets you index the documents (.xslx, .xls, .csv) using luecene indexing provided by llama_index and predict the prompts/ queries using LLMPredictor provided by Open AI's model text-davinici-003

Technical Specifications

• DRF (Django Rest Framework)
• Python 3.9 +
• Any IDE for Python development (PyCharm)

OPENAI_API_KEY (Linux/ Mac Setup)

• Open the .bash_profile in any editor of your choice.
• Then edit it by adding the below line export OPENAI_API_KEY=<Key Value> at the end of the file.
• If there is no API key available, then you can create it from API Key, check the Usage and expiry date for the Open API Keys, if the Key is not valid, they will be no response provided and it will lead to RateLimitErrors.
• Once the above line is added to the .bash_profile file, then load it using the command source ~/.bash_profile and restart the terminal session.

OPENAI_API_KEY (Windows Setup)

• Create a new environment variable from the Advanced System Settings.
• The name of the variable should be OPENAI_API_KEY, create it as a system/ user environment variable based on the administrator access and privileges.

Setup Process

• Internet Connection (for resolving dependencies & accessing Github APIs)
• Checkout the code or download a zip file.
• cd to the directory search-docs
• pip install -r requirement.txt to download all the dependencies.
• Once the packages are successfully installed, cd to the root directory where you can find manage.py
• By default a superuser is created with kathi/ihtak as username/password.
• To create a new superuser, execute python manage.py createsuperuser and follow the prompts.
• To use this newly created superuser to get the access tokens, change the username and password to the chosen values in the script of index.html
• Delete all contents inside the data folder, if you want only specific data to be indexed and the data which is already available is the sample data.
• To start the Server run, python manage.py runserver, this starts the server on the Port 8000 which is a default port and the home page is Home

Execution of prompts

• Open the Homepage and upload the document.
• After the document is successfully uploaded, it will then be indexed and the Question input will be enabled.
• Ask for any question to get the response and check for the Examples section
• Don't rely completely on the Response, it's mostly accurate but, it is prone to errors. Please be wary of taking decisions solely on the response without a proper check.

Examples

• What is this about? (You would get the response as a summary of what the uploaded document is about.)
• What are the colums of this dataset? (You would get the columns of the dataset)
• What is the mean/ average value of a <your_column>? (You would get the average value of that specific column)
• You could use the techniques involved in Prompt engineering to modify the prompts provided to get a more better and accurate responses which you could rely on.

References

• Prompt engineering for Developers
• Source Github for code inspiration.
• Youtube is a good source for learning LangChains, Search for LangChains or llama_index with Open AI

Applications

• llama_index can index many things by default Out of the box from the package, for example, it can take the youtube URL and then read the transcripts, we can then query/ prompt about the video. Many other applications can be thought of to build a chatbot powered by Open AI.