Projects and Case Studies
Publicly shareable development experience.
I reorganized verifiable materials and source-code history without exposing confidential information. The work spans research development, field measurement, AI environments, and legacy-asset repair.
Information Retrieval / Agriculture / SDD
Concept search over more than 200,000 Japanese agricultural documents
Research and implementation of a search system for large agriculture, forestry, and fishery document collections using Semi-Discrete Matrix Decomposition. It handled vector-space models, LSI, SVD comparison, and memory reduction by SDD.
- Public outputs
- Papers, slides, accuracy-improvement notes, SDD concept-search materials
- Technologies
- Information retrieval, VSM, LSI, SVD, SDD, Java, web search, agriculture corpus
- Current use
- AI search, RAG, internal knowledge search, specialist document search
Java / XML / AgrInfo
AgrInfo search client and XML configuration system
Built an agricultural information system including a Java/SWT search client, XML/XSD/JAXB settings, and communication with a search server.
- Public outputs
- AgrInfo/SDD Java source, XSD, setting XML, SWT client structure
- Technologies
- Java, SWT, XML, XSD, JAXB, socket communication, search UI
- Current use
- Legacy Java maintenance, research-tool rebuilds, migration to web management UI
Android / Web / Field DX
Barcode measurement system phase 2
Prepared requirements and detailed specifications to evolve a phase-1 PoC into a field-ready system with Android devices, Bluetooth measurement equipment, Room storage, sync, web management, Supabase Auth, InfluxDB/Grafana visualization, and CSV/JSON output.
- Public outputs
- Requirement docs, detailed specs, screen design, API/DB policy, progress materials
- Technologies
- Android, HID/SPP/BLE, Room, Supabase, InfluxDB, Grafana, CSV, JSON
- Current use
- Farm trials, factory/research measurement management, terminal input and web visualization
C / AI / Windows App
C neural-network program repair and GUI conversion
Fixed memory issues when expanding an existing C neural-network program, converted images to training samples, and created a Windows app to check learning, prediction, logs, and error transitions.
- Public outputs
- Repair report, Visual Studio project, Python image-conversion script, verification materials
- Technologies
- C, Visual Studio, Windows GUI, Python, image conversion, memory management
- Current use
- Old research-code repair, C/C++ asset reuse, operationalizing existing AI programs
Flower / YOLO / Docker
Flower + YOLO environment for federated learning
Investigated and built a Flower federated-learning environment so image data from multiple locations could be used without collecting it in one place.
- Public outputs
- Dockerfile, compose.yaml, client_app.py, yolo_task.py, execution procedures
- Technologies
- Flower, YOLO, Docker, Hyper-V, Python, COCO, federated learning
- Current use
- AI experiments with distributed data, research PoCs, environment setup and procedures
Business Systems / Migration
Search engine, web, and securities-system development and migration
Experience in search-engine development and migration, web systems, securities-system migration, server setup, testing, and maintenance. The ability to read old specifications and existing code is the base of current AI work.
- Public outputs
- Career materials, project detail sheets, progress and design materials
- Technologies
- C, Java, UNIX/Linux, Shell, Tcl/Tk, Tomcat, DB, test automation
- Current use
- Legacy investigation, migration, maintenance improvement, safe system modification