Enhancing Ghanaian Political Tweet Engagement



 Achieved a 27% perplexity decrease through unsupervised training(masked language modelling) using Distil RoBERTa to adapt to the Ghanaian political domain.

 Leveraging the domain-adapted model, developed a predictive model for predicting tweet engagement.

 Finetuned the GPT-3 model to optimize tweet virality by generating engaging variations for low-engagement tweets, ultimately yielding a 25% improvement in overall engagement metrics.

 Strategically deployed models on Google Cloud Platform: Utilized Cloud Run for serverless architecture, Employed Docker containers for efficient deployment, Collaborated cross-functionally for successful deployment and ensure ongoing stability.