A longitudinal record of technical trajectory, ontology engineering, and distributed systems orchestration—documenting academic research at JAIST and professional work in high-performance environments.
Research Assistant — Galactic Archaeology & ML
Architecting an optimized HDBSCAN clustering pipeline for reconstructing galactic merger histories. Implementing AutoGluon-based supervised learning to isolate faint stellar structures within high-dimensional chemodynamical datasets.
Joint Research — Ontology Engineering
Engineered semantic extraction pipelines for ADS safety testing. Validated ontological models on the KAGAYAKI HPC cluster.
Teaching Assistant — Computer Science
Mentored 50+ students per class in Python and Assembly. Authored technical documentation and automated workspace setup guides.
DevOps Contractor — Systems Infrastructure
Architected automated grading systems and microservice orchestration layers for the Computer Science Department.
Informal observations & technical field notes
Scaling PBS orchestration on the Kagayaki HPC cluster requires a strict hierarchical approach to job script generation. We implemented a BASH-based factory pattern that dynamically injects resource requirements based on node-pool availability. This reduces wait-time by 22% and prevents the common 'stuck-in-queue' deadlock seen in monolithic submissions.
Our research at JAIST focused on the automated extraction of structured knowledge from large-scale unstructured datasets. By leveraging Python-based analysis pipelines, we successfully mapped raw textual data to formal ontological classes, enabling high-fidelity reasoning over complex domain models.
We implemented a hierarchical permission model that decouples object storage from access logic. Using Neon PostgreSQL as a centralized state authority, the system evaluates incoming ACCESS_REQUEST pulses against predefined user tiers (Guest/Admin/Root). Validated requests trigger a just-in-time permission grant on the GCS bucket layer, ensuring that terabyte-scale research data remains encrypted at rest while maintaining sub-second authorization latency for verified auditors.
We engineered a robust word search and dictionary resolution engine leveraging a multi-tiered fallback architecture. The system models a Word as a strict ontological entity, normalizing its lexical format, grammatical category, and source provenance across heterogeneous data streams. The retrieval pipeline minimizes latency by using an O(1) Redis L1 cache and PostgreSQL L2 index before falling back to sequential remote API resolution. Client-side gameplay is validated instantly through Prefix Tree (Trie) backtracking algorithms.