Jia Son

Automating production of university-entrance English study material 2026.06 ~ now

• Built a pipeline that parses sentence structure and generates translations and commentary
• Built a tool that extracts headwords and classifies vocabulary
• Moved the work to a review-first workflow by generating drafts automatically, cutting turnaround time by 80%

R&D on automated legal document analysis for lawyers 2026.04 ~ 2026.05

• Designed and implemented an AI workflow: statute search → evidence extraction → answer generation
• Cut input token cost by 75% with disk and context caching that avoids repeat LLM calls
• Added LangFuse-based tracking of token usage and cost per request and per model
• Ported a minimal Korean Law MCP server and put it into use

Improving hybrid search quality (rank fusion) over legal documents 2026.04 ~ 2026.05

• Built a quantitative evaluation harness against a BM25 baseline
• Optimized rank fusion weights, doubling top-result relevance (RBO@10 up 67%)
• Cut latency by over 20% with no loss in vector search quality

Designed and built a full-stack web app for brain tumor AI segmentation and 3D reconstruction 2025.08 ~ 2026.02

• Designed the UI and UX in Figma from the research proposal
• Redesigned an offline batch codebase into an async API service, cutting wait time by over 80%
• Implemented the front end (HTML/CSS, JavaScript) and back end (FastAPI, Celery, Redis)

FigmaHTML/CSSBootstrapJavaScriptFastAPICeleryRedis

MICCAI BraTS-Path, “Assessing the Heterogeneous Histologic Landscape of Glioma” (3rd place) 2025.06 ~ 2025.08

• LoRA fine-tuning of a foundation model, data-parallel training, and model ensembling; wrote the paper, served as a peer reviewer, and presented the results at the workshop

PyTorchHuggingFaceW&BDockerGit

Refactored the brain tumor AI segmentation and 3D reconstruction pipeline 2025.05 ~ 2025.10

• Introduced structured logging, cutting debugging time by about 30%
• Added exception handling to improve pipeline stability and error recovery
• Modularized shared logic, improving reuse and maintainability

PythonDocker

R&D on brain tumor AI segmentation and 3D reconstruction 2023.12 ~ 2025.02

• Researched an assisted labeling interface and a high-precision data pipeline for medical image analysis
• Benchmarked and validated MR and CT segmentation and 3D reconstruction tools
• Built an automated pipeline that cut processing time by 80% against manual work

PythonPyTorchMONAI3D SlicerDocker

Built medical AI training datasets and designed the annotation scheme 2023.09 ~ 2025.02

• Automated 3D label extraction from metastasis MR scans, cutting processing time by 70% against manual work
• Built a high-quality dataset of 31,000 traumatic brain injury CT scans (90% of the total)
• Ran data quality control for research on VLM-generated CT reports
• Introduced AI-assisted labeling that cut hemorrhage CT annotation time by 50% and improved consistency between readers

Python3D SlicerPyTorchDocker

Built and operated medical AI research infrastructure 2023.09 ~ 2025.02

• Set up remote access to internal research resources, with access control
• Ran NAS-based storage and managed research data
• Contributed over 80% of the R&D knowledge base

Synology NASTailscaleNgrokUbuntu Linux3D SlicerNotionGit

Platform design and standard SW development artifacts for the Delivery Service Mutual Aid Association 2023.03 ~ 2023.06

• Designed the ERD and entity relationships for insurance incidents and claims
• Designed and specified API endpoints from the requirements

ReactSpringRDB

Research on intelligent AI components for a SaaS finance system 2022.08 ~ 2023.06

• Modeled the finance domain across assets, liabilities, revenue, and expenses
• Researched a financial product recommender driven by a user’s financial position

PythonPyTorchRDB

R&D on information services linked to a robo-advisor 2022.05 ~ 2022.08

• Compiled a domain glossary of financial indicators and terminology
• Designed the data interface and the LLM prompting
• Built a prototype service for LLM-based stock volatility analysis (OpenAI GPT-3, FastAPI, MongoDB)

PythonPyTorchFastAPIMongoDBOpenAI API

Mentoring, curriculum design, and teaching on project-based bootcamps (223 hours)

• Kaggle mentor, Yeardream School cohort 5 (2025)
• Project coach, Elice Track AI Campus (2026)

AI upskilling program for corporate teams (52 hours)

• AI adoption training for executives at S Electronics and S Electronics Logitech
• PBL mentoring on marketing automation for a bio-healthcare company

Mentoring on a data analysis and AI bootcamp (101 hours)

• Weekly study guidance and career mentoring
• Reviewed assignments and gave written feedback
• Advised multimodal projects and gave per-team feedback

Guest lectures on deep learning for medical AI (8 hours) and career mentoring (10 hours)

• Lectures on foundation models, computer vision, NLP, and LLM theory
• Wrote and taught hands-on medical AI exercises
• Career advising for learners nearing graduation

2025 BoostCamp, mentoring on the AI Tech cohort 8 NLP track (60 hours)

• Graded AI engineering assignments and reviewed deep learning code
• Coached teams through ODQA and LLM competitions on college entrance exam questions
• Advised agent development projects

2024 BoostClass, coaching on the KDC AI engineering fundamentals program (48 hours)

• Answered questions in real time through the platform
• Graded Python deep learning assignments and reviewed code
• Ran a Notion page for my assigned learners
• Built and published a web page with model answers