TEGD_thirdai_hackathon

A neural DB based google drive search engine

Installation

1. Wispher

!git clone https://github.com/ggerganov/whisper.cpp.git  
%cd /content/whisper.cpp/models  
!bash download-ggml-model.sh base.en  
%cd ..  
!make   

2. ffmpeg (apt)

Requirements

pydub 
thirdai
thirdai[neural_db]
langchain
openai
paper-qa
requests
bs4
pandas

Directory Structure

• book_web_Scrapping:

  To Download the books from the web (NCERT)

• Thirdai_gdrive_engine_TEGD.ipynb

  main script to run the search engine

Usage

To run the search engine:

• Open the notebook with google colab
• mount the drive by executing the very first cell

from google.colab import drive
drive.mount('/content/drive')

NOTE:

We are having a user input option in the notebook, as the extraction and training of audio file is time consuming.

Please give input as y to the cell if you want to train your G-Drive's audio files.

mp3_train = False
print("Do you want to fetch and train mp3 files too?")
print("It will take a lot of time. You can skip it for now")
if 'y' in input("Press 'y' to extract audio files else press 'n'").lower():
  mp3_train = True

By running the notebook Thirdai_gdrive_engine_TEGD.ipynb

• The drive will be mounted to the notebook
• Get all the files in your G-Drive

Currently we are working on the pdfs, docx, mp3 and wav files. The other file format like all source code file and mp4 file will be added soon.

• For processing audio files

  • Download ggml-whisper which is a really fast version of whisper written in C/C++
  • Convert an audio file with any format to wav (the only format currently supported by ggml-whisper)
  • Convert the transcription to csv and save for model training later

• We are storing a map for each audio_csv file to its original localtion in the drive. So that when we query from the engine, we can redirect the output to original location.

• Get the General_QnA model from the bazaar

• PREPARE INSERTABLE DOCS

  • from pdfs
  • docx
  • mp3 (created earlier)

• SEARCH THE QUERY (from user input), and you will get the exact location of the file in the drive.

DATA Link

https://drive.google.com/drive/folders/1tNJ-r6-URDAcU0VOZJ_h50hs4PrO5EkN?usp=sharing

Try Our Demo

DEMO

Note: We are not submitting the model, as our main Moto is to keep data private. You can just train your model on your own personal data.

#POCKETLLM