ASKPDFANYTHING

In the life of a student, time is of crucial importance yet, a good amount of time is spent in searching for answers rather than being able to understand those. To bridge this gap, I made this project

Comment

Inspiration

To the countless hours I wasted just to search and verify a fact/formula

As a student, I strongly feel that time must be spent learning what counts rather than scraping pdfs to verify one fact that is needed for an assignment and these countless one facts that complete the assignment. So, to tackle this problem I made AskPdfAnything it leverages the power of LLMs to answer the users questions acc. to the given document/pdf.

What it does

It takes in a pdf file as an input, and reads it using LangChain to convert it to their equivalent word embedddings then, LLMs come into play (using Ollama) and handle the doc and the user query to answer the question.

How we built it

Model was built using LightningAI VMs which are free for students
I have used LangChain for reading pdfs
For creating vector-embeddings I have used Nomic-Text
The vector-embedding is made of the input query and the doc
The vec-embedding is passed onto the Mistral decoder to return the relevant ans

Challenges we ran into

While this model can be run on a standard laptop having 16gb ram and 8 cores, it takes about 7-8 minutes per query so, using the model with a GPU is strongly recommended which led me to use lighning AI to tackle my problem
Understanding the model architecture to be put in place
Making the app interface using Gradio

Accomplishments that we're proud of

Its able to answer question asked wrt document with a decent accuracy and to the point

What we learned

TO leverage LLMs in real life
TO make a neat gradio interface
How to use LangChain

What's next for ASKPDFANYTHING

Improving the truthfulness score of the model
Testing it
Changing underlying LLMs
Deploy for the World to Use

Built With

Share this project:

Updates