Emotibot: BERT sentiment analysis with visual feedback

The experiment

Emotibot explores how a trained language model can turn free-form text into an understandable sentiment result. A BERT model trained on a Twitter dataset classifies the input, and the interface maps the result to a corresponding emoji.

System design

• React and Next.js provide the user interface.
• Flask exposes the model through a small API.
• BERT performs multi-label sentiment classification.
• Docker keeps the model service and application environment reproducible.

Engineering focus

The main work was preparing the training data, evaluating closely related sentiment labels, and connecting a Python model service to a responsive web interface. Containerization made it easier to run the same model stack during development and deployment.

Result

The project provides a compact demonstration of an end-to-end machine-learning product: training, inference, an API boundary, deployment packaging, and a user-facing explanation of the output.