Multi-Lingual Chatbots

Key Technologies: Python, NLP, Active Learning, Flask, Tensorflow

The Challenge

Our client wanted to expand their chatbot capabilities to multiple European countries, and improve upon their chatbot’s understanding of human conversation. Finally, they wanted the models to improve continuously.

The Solution

Adinkra co-developed an active learning solution to continuously train the models. We then used multilingual BERT-based checkpoints to fine tune for our application’s specific entity recognition, sentiment, and topic analysis needs.

The Impact

Our client deployed their product to multiple countries in Europe successfully. Their internal labeling team decreased data labeling efforts by 30%.