https://github.com/aml-hassan-abd-el-hamid/the-finder

ML web-based system that can find similar products based on user inputs
https://github.com/aml-hassan-abd-el-hamid/the-finder

cohere machine-learning nlp semantic-search streamlit

Last synced: 4 days ago
JSON representation

ML web-based system that can find similar products based on user inputs

• Host: GitHub
• URL: https://github.com/aml-hassan-abd-el-hamid/the-finder
• Owner: Aml-Hassan-Abd-El-hamid
• Created: 2023-12-18T16:40:19.000Z (about 1 year ago)
• Default Branch: main
• Last Pushed: 2023-12-21T15:56:49.000Z (about 1 year ago)
• Last Synced: 2025-01-20T22:53:56.283Z (16 days ago)
• Topics: cohere, machine-learning, nlp, semantic-search, streamlit
• Language: Python
• Homepage: https://the-finder-qmzffewnwa5tvvuvappepfq.streamlit.app/
• Size: 23.4 KB
• Stars: 3
• Watchers: 2
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

        
# The Finder

ML-web-based system that can find similar products based on user inputs

That's actually a part of an interview process, I was given 7 hours to do exactly the following:

```

- Build an ML-web-based system that can find similar products based on user inputs, then deploy it. So, we have only one textbox for product input and another one “ie.. label” for model outputs.

You can focus on Arabic-written products only.

Plz, be cautious of similar items such as “هاتف” and “غطاء هاتف”, so each of them shouldn’t appear in the other one’s search.

You can use the attached JSON file “Black Friday” for validation.

```

I now have 5.5 hours left -I spent the first 90 minutes planning the work :) - so let's go

I'm done with the initial look of the app, now I have about 4.5 hours, and I'm faced with the winning question, what model should I use to match the product name to one of the products that I have in the validation list of product?


I wasn't provided with training data, computational power or time :) so the best option that I know that I have is a pre-trained model from the amazing HuggingFace 🤗.


Which model should I use in particular? Will that thing is what I'm about to discover, for now, I just know that the main task that this model has to accomplish and the main characteristic that I have in my head currently is that this model has to know how to deal with Arabic language text.

I still have 90 minutes to the deadline but I think I'm done :) through the last 3 hours, I figured out that HuggingFace might not be the most suitable or fastest option, especially with the Arabic language so I turned to Cohere, it gave very good results on the validation set.


I used Cohere's multilingual-22–12 model, it's an amazing model that supports over 100 languages, and it's considered the industry’s first multilingual text understanding model to support such no.of languages! you can know more about that model by taking a look [here](https://txt.cohere.com/multilingual/).

Here's a link to what the version that I sent looked like: https://the-finder-qmzffewnwa5tvvuvappepfq.streamlit.app/

Here's a list of things that I intend to do in the future to turn that into a more CV-friendly project

ToDo:

  1- Improve website look


  2- Get all the products


  3- Get more product data to train on.


  4- Fine-tune the model of the data.