Senior Nlp Researcher
Current- Currently, applying LLMs technology: automatic summary of conversations, intelligent text search based on text embeddings...- Designed, trained and evaluated the NLP models of Aircall's transcription engine: models for addition of punctuation marks, capitalization, and rescoring of alternative transcriptions, in English (RNNs trained in PyTorch with ~100M tokens)- Fine-tuned BERT for NER and emotion identification in English, French and Spanish (with HuggingFace libraries and ~2M tokens)- Exploration of large collections of conversations (~100K): clustering of texts, extraction of relevant terms, identification of problematic conversations with sentiment analysis