VALORANT Round Outcome Prediction
With NAIST, I used TimeSformer to predict VALORANT round outcomes from minimap video.
Python · Deep Learning · Computer Vision · TimeSformer · OpenCV
Nirai Hayakawa
Full-stack / AI infrastructure
I study computer vision, LLMs, and game AI at CU Boulder. In Japan, I build AI products and the infrastructure behind them.
Featured Projects
Texture Platform Automation
Automated texture packaging for Unity and Unreal Engine with Python and LLM-assisted validation.
Python · LLM · Automation · Productivity Tool
LLM Bias in Algorithm Evolution
Ongoing research into conservative local optimization by LLMs in ShinkaEvolve and OpenEvolve.
LLM · Research · Algorithms · Heuristics · Bias Analysis
Speech-to-Speech LLM Agent
Independently built a CLINKS Corporation proof of concept for speech-in and speech-out LLM interaction.
Python · LLM · Speech-to-Speech · AI Agent · R&D
AI Agent Evaluation Framework
At Not A Hotel Inc., I am researching and implementing repeatable AI-agent evaluation with LLM as a Judge.
Python · LLM · AI Evaluation · Research · Automation
Experience
-
Not A Hotel Inc.
Software Engineer Intern
- I work on the backend for Kevin, NOT A HOTEL's AI concierge, and the shared API platform used across the company.
- I research and implement ways to evaluate AI agents.
- I use LLM-as-a-judge methods to automate parts of that evaluation.
- I build evaluation tooling that teams can reuse across AI products.
- I also work on digital transformation in architecture.
- I handle several computer-vision tasks.
-
CyberAgent, Kiwami AI Platform Team
Full-stack Engineering Intern
- Built a PoC that split advertising videos into scenes.
- Improved the performance of a quality-assurance program for generated ads.
- Built QA deployments that reproduced each branch's Terraform diff.
- Added end-to-end tests.
- Implemented feature flags that static analysis could resolve at build time.
-
CLINKS Corporation
R&D Product Developer Intern
- Independently built a PoC from initial design through implementation.
- Built a speech-to-speech LLM agent system.
- Prepared the demo for public release and stakeholder review.
-
Bird Fab Studio
Software Developer (Part-time)
- Built an automated texture-packaging system with Python and LLM integration.
- Reduced review failures to nearly zero and added automatic proofreading and translation.
- Created documentation for the texture platform.
- Built internal tools, including native applications.
About
My published computer-vision work predicts VALORANT round outcomes from minimap video. I have also built voice agents, texture tooling, and AI evaluation systems.
At CU Boulder, I study how LLMs behave inside algorithm-evolution frameworks for NP-hard problems.
Latest Writing
A Gopher's TSKaigi 2026 Student Travel Support Retrospective
A retrospective on attending TSKaigi 2026 through student travel support, from TypeScript backends and type inference to sponsor booths, scholarship lunch, OST, and the party.
- conference
- typescript
- tskaigi
- backend
- student
Building a Scalable Backend for an LLM Chatbot During a Two-Week Internship at GMO Internet Group
During a two-week internship in GMO Internet Group's ML/WEB course, I joined a five-person team building an MVP and took charge of the backend for an LLM chatbot. This is a retrospective on the requirements, technology choices, design decisions, failures, and lessons learned.
- internship
- backend
- llm
- typescript
- trpc
Publications
Ongoing Research
Ongoing Projects
Certifications
- AWS Cloud Solutions Architect Professional
- Google Cloud Cybersecurity Professional
- Google Data Analytics Specialization
- DeepLearning.ai Deep Learning Specialization
- IBM Generative AI Engineering Professional
- Microsoft AI & ML Engineering Professional
- IPA Applied Information Technology Engineer
Achievements
- AtCoder Rating: 1252 (Light Blue)
- Japanese Olympiad in Informatics: Finalist
Say hello.
For engineering work or research collaborations, email me. LinkedIn works too.