An archive of my talks at conferences and meetups on topics like Machine Learning, LLMs, and building AI-powered products.

Beyond LMArena: 3 Practical Insights for Custom LLM Benchmarks

Tokyo Tech Leads Circle | Oct 2025 | Slides

On “interviewing” LLMs on your actual tasks, not just public benchmarks. Sharing three practical insights for building your own benchmark: starting small, keeping the eval pipeline simple for debugging, and using pairwise statistical tests to make confident model choices.

Intuitive Jargon-Free Mental Model for MCP

Tokyo AI | Jun 2025 | Slides

On finally getting why MCP exists. Grasping the core concept with a browser analogy, seeing how it modifies system prompts, and watching it all work in live demos with Roo Code.

AI-powered Automatic Replies in Customer Support

PyCon JP | Sep 2024 | YouTube | Slides | Details

On building a precision-focused auto-reply system for customer support. Detailing the process of fine-tuning a transformer for Japanese text classification, ensuring a good user experience, and measuring its real-world business impact.

AI for Efficient Routing of Customer Inquiries at Mercari

PyCon APAC | Oct 2023 | YouTube | Slides | Details

On translating the business need for better customer inquiry routing into an ML problem. Going from simple keyword matching to fine-tuning a transformer for production use, and its real-world impact on support operations.

---