Jason Martin

Key contributor in developing a sentiment analysis engine using RNNs and transformers (BERT, XLNet), achieving 74% sentiment classification accuracy and enabling more effective tracking of customer experience trends. Designed a retraining pipeline using user feedback to improve model robustness, reducing false negatives in negative sentiment detection. Built a churn prediction model employing XGBoost and Random Forest classifiers to identify high-risk customers and supported Customer Success teams that reduced churn rates by 17%. Developed a scalable WordCloud generation engine with advanced n-gram modeling and lemmatization, replacing third-party solutions and enhancing multi-language support for 13 additional languages. Constructed modular NLP pipelines for text preprocessing and visualization, enabling automated enterprise survey insights. Partnered with Analyze and Engineering teams to deliver customized embedded analytics components supporting client-facing survey insights.

Spearheaded the integration of Pearl GPT with human-in-the-loop validation workflows, leveraging GPT transformer models to enhance AI response quality, resulting in a 42% improvement in trust metrics. Designed and prototyped the Expert Conversation Wizard using prompt engineering and transformer-based NLP techniques that boosted expert response clarity and increased user satisfaction by 25%. Built and managed a customer sentiment analysis pipeline with BERT and RoBERTa classifiers integrated into real-time processing systems, improving customer support outcomes by 18%. Explored and implemented Retrieval-Augmented Generation (RAG) pipelines using FAISS and transformer models to enhance factual accuracy and contextual relevance of AI-generated responses. Collaborated closely with product, engineering, and UX teams to align data science solutions with product initiatives, supporting agile delivery and iterative improvements. Provided mentorship and technical guidance to junior team member

Developed a sentiment analysis engine using RNNs and transformers (BERT, XLNet), raising classification accuracy to 74%. Designed a retraining pipeline using user feedback to reduce false negatives for negative sentiment detection. Built a churn prediction model with XGBoost and Random Forest classifiers, reducing churn rates by 17%. Developed a scalable WordCloud generation engine supporting 13 additional languages, improving performance and replacing third-party solutions. Constructed modular NLP pipelines for scalable automated survey insights. Partnered with Analyze and Engineering teams to deliver customized analytics components for embedded, client-facing survey insights.