Manato Ito
Software Engineer | LLM Specialist
Kawagoe,11
Summary
Thorough team contributor with strong organizational capabilities. Experienced in handling numerous projects at once while ensuring accuracy. Effective at prioritizing tasks and meeting deadlines.
Overview
4
4
years of professional experience
Work History
Grand Prize Winner, LLM Competition 2025
Matsuo-Iwasawa Lab, The University of Tokyo
05.2025 - 09.2025
- Achieved 1st place overall by developing a high-reasoning LLM that reached 18.58% accuracy on Humanity's Last Exam (HLE), a +0.79% improvement over the base model and a top-tier result among open-source LLMs.
- Executed large-scale fine-tuning of the Qwen3-235B-A22B-Thinking-2507 model using GRPO (Group Relative Policy Optimization), optimizing the model’s performance for complex, PhD-level reasoning tasks.
- Engineered a RAG (Retrieval-Augmented Generation) pipeline that indexed high-quality academic papers from arXiv, significantly enhancing the model's ability to provide evidence-based answers to advanced scientific queries.
- Spearheaded the technical research and validation of state-of-the-art fine-tuning techniques, implementing a "tournament-style" evaluation framework to iteratively select and refine the most effective training methodologies.
- Orchestrated the LLM development roadmap, managing end-to-end schedules from initial research to final model deployment, ensuring technical feasibility and milestone adherence within a team environment.
Category Winner, MWSCup 2025 (Static Analysis))
MWS Organizing Committee, Special Interest Group on Computer Security (CSEC), Information Processing Society of Japan (IPSJ)
10.2025 - 10.2025
- Won 1st place in the Static Analysis category, scoring 24.60/25.0 points and outperforming the runner-up by a significant margin of 2.60 points in a nationwide competition with over 150 participants.
- Developed a high-precision malware classification model by implementing an ensemble learning approach (Random Forest and LightGBM), achieving robust detection accuracy despite a limited training dataset.
- Optimized team performance through systematic analysis of past competition problems, leading a strategic initiative to assign roles based on individual technical strengths, which maximized the team's overall output.
- Conducted in-depth static analysis on malicious samples, utilizing Python scripts to extract critical features for the machine learning pipeline.
Cram School Instructor / Tutor
WASEDA ACADEMY CO., LTD.
05.2022 - Current
- Transformed student academic performance to exceed competitive entrance exam thresholds by executing a data-driven, multi-faceted intervention strategy following critical stakeholder feedback.
- Identified and addressed systemic learning gaps by conducting 1-on-1 diagnostic interviews for all students, creating personalized "weakness maps" and tailored study plans.
- Established a collaborative support ecosystem between the school and parents, initiating regular progress reporting and home-study synchronization that increased student engagement.
- Achieved high success rates in admissions for my assigned subjects and converted parental dissatisfaction into long-term trust and positive testimonials through consistent accountability and results.
- Demonstrated strong ownership and communication skills over 4 years, effectively bridging the gap between complex curriculum requirements and individual student needs.
Education
Master of Science - Computer Science
Waseda University
Tokyo, Japan 04.2001 -
Bachelor of Science - Computer Science
Waseda University
Tokyo, Japan 04.2001 -
Skills
Python
PyTorch, TensorFlow, Hugging Face Transformers
Large-scale LLM Fine-tuning (GRPO), RAG (Retrieval-Augmented Generation)
GitHub
JDLA E-Certificate 2025#2 (Deep Learning Specialist)
English (Business Level)
Japanese (Native)
Timeline
Category Winner, MWSCup 2025 (Static Analysis))
MWS Organizing Committee, Special Interest Group on Computer Security (CSEC), Information Processing Society of Japan (IPSJ)
10.2025 - 10.2025
Grand Prize Winner, LLM Competition 2025
Matsuo-Iwasawa Lab, The University of Tokyo
05.2025 - 09.2025
Cram School Instructor / Tutor
WASEDA ACADEMY CO., LTD.
05.2022 - Current
Master of Science - Computer Science
Waseda University
04.2001 -
Bachelor of Science - Computer Science
Waseda University
04.2001 -