Tweet Charity Recommender - Data_R_Us Team - IM09

• 

• 

  Technical Flowchart

Inspiration

We felt that currently there is so much conflict occurring and it is not always easy to find reliable sources to donate to especially on twitter(X) and other platforms. We set about to create an application that could resolve this issue.

What it does

Takes a tweet if relevant, and returns a list of charities URLs that are relevant to the tweet subject.

How we built it

Used LLMs to classify unlabelled data, which we then trained an XGBoost model from via vector embedding. Deployed this model alongside NLP on a Flask backend to classify tweets from our twitter emulating frontend.

Challenges we ran into

Hosting Llama on Z IBM Stack, as it requires extensive memory.

Accomplishments that we're proud of

Generation of labeled data using LLMs. Using this data to train an accurate XGBoost model (89% weight average Precision, Recall and F1_Score). Web scraping URLs of relevant charities.

What we learned

IBM Z stack deployment. Full stack development. Team Work. LLM Inference.

What's next for Data_R_Us Team - IM09

Integrate with Twitter through API or approach twitter/X and ask for integration into application/framework.

Built With

• embedding
• flask
• flutter
• llama
• llms
• ml
• natural-language-processing
• python
• webscrape
• zos