・Developing and training Speech LLM with novel architecture specialized in customer calls.・Developing custom data preprocessing, speech tokenization and post-training methods for Japanese speech.・Developed full-stack AI solutions for five external customers such as patent analysis tools using RAG and a system for synthetic abnormal data generation through training custom diffusion models using Pytorch.

• Trained and deployed machine learning models for scoring English proficiency tests impacting 420K learners.• Improved the model accuracy by improving data preprocessing and fine-tuning BERT models using scikit-learn, HuggingFace Transformers and AWS SageMaker.• Automated model deployment process across two repositories using GitLab CI/CD and GitHub Actions.

• Developed an internal tool using Rhino 3D and Grasshopper, which calculates unfolding 3D animation and 2D diagrams of various origami structures based on multiple input parameters.• Enabled the CTO to prototype and test the structure digitally on software, eliminating the need for manual creation of physical origami structures.