Skip to content
Projects

AICA Government AI Datacenter Program

Government AI datacenter GPU infrastructure program. Led end-to-end from proposal writing and research execution to quarterly reporting and outstanding performance recognition.

2024.01 — 2026.03 BHSN AI Engineer (Program Lead)
GPU Infrastructure Government Program H100 Outstanding Performance

Overview

Led the full lifecycle of the AICA (AI Innovation Center Alliance) government AI datacenter program — from proposal writing → infrastructure setup → research execution → quarterly reporting → outstanding performance recognition.

End-to-End Program Operations

Proposal Writing & Application

  • Authored the legal AI GPU compute program proposal
  • Designed research goals, expected outcomes, timeline, and budget
  • 2024 initial application and 2026 re-application preparation

Infrastructure Setup

  • Resolved H100×8 node resource allocation, handled offline paperwork
  • Directly configured firewall/NAT settings
  • Designed architecture connecting AICA vLLM serving to internal aihub/LiteLLM chain

Research Execution

  • 2024: Legal-domain LLM training, semantic search model, document structure analysis (H100×8, A100)
  • 2025 H1: Multimodal legal reasoning LLM Agent (H100×8)
  • 2025 H2: Legal-domain LLM-as-a-Judge auto-evaluation system (A100×8)

Performance Reporting

  • Authored and submitted quarterly performance reports
  • Organized commercialized services sections and compiled metrics
  • Coordinated team role assignments and managed final submissions

Outstanding Performance Recognition

  • 2026 Q1: Selected for 2025 Outstanding Achievement Collection — research recognized by government evaluation

Key Research Achievements

  • Legal semantic search model: 90%+ accuracy, 70% latency reduction
  • Korean contract review accuracy 22% ahead of OpenAI models
  • Document structure analysis F1 score 92%
  • 3 government PoC projects: National Human Rights Commission, KARSO, KEPCO KPS

Tech Stack

H100 GPU, A100 GPU, PyTorch, Transformers, vLLM, LiteLLM, FAISS, FastAPI, Python