Skip to content
View soeque1's full-sized avatar

Organizations

@ski-net @tdd-master @HephaestusProject @EleutherAI

Block or report soeque1

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
soeque1/README.md

Senior LLM/AI Engineer | Full-stack LLM: Pre-training → RLHF/DPO/GRPO → Serving → Evaluation | Apache MXNet Contributor | Ex-BHSN(CAIO) · NAVER · SKT

I build GenAI systems that work in the real world — not just in demos. Over the past 10+ years, I've designed and shipped production-grade LLM systems, Agentic workflows, and enterprise AI platforms serving Fortune Global 500 companies and top-tier law firms in Asia.As Chief AI Officer at BHSN, I led the end-to-end development of Korea's leading Legal AI platform — from domain-adaptive LLM pre-training (allibee astro) and DPO-based alignment with legal citation reward functions, to graph-structured Legal Agents and RAG pipelines deployed on Public/Private Cloud and On-premise. The platform earned ISO 27001, ISO 27017, and CSAP certification — meeting the strictest enterprise security and compliance standards in a highly regulated domain. BHSN was selected for Google for Startups Cloud Program (2024), featured on Google Cloud official blog alongside Mistral and Pinecone.

Shipped Legal OCR product to AWS Marketplace.Before BHSN, I spent years at NAVER and SK Telecom building the foundation:

— Built core NLP features for ClovaNote (500K+ MAU) and Summarization API for NAVER WORKS (4.5M+ global users) — Designed HyperCLOVA X Skill System at NAVER Cloud (production Tool Use / Agentic architecture, analogous to today's MCP) — Alignment learning for KoGPT3 at SKT — 2nd Place at DSTC8, the international dialogue systems challenge (IEEE Access 2021) — Contributed Korean evaluation benchmarks to EleutherAI lm-evaluation-harness and to Apache MXNet core; published on the official Apache MXNet Medium blogWhat I care about most is the gap between "impressive demo" and "running in production at scale." That's where I live.

📌 Expertise: LLM pre-training & fine-tuning · DPO / RLHF / GRPO · Distributed training (FSDP, vLLM, FlashAttention) · LLM evaluation (Promptfoo, LLM-as-Judge) · RAG & vector search · Agentic workflow design · Multi-cloud deployment · MLOps · Regulated environments (Legal, Financial, Public Sector)

📌 AWS Startup Jungle Government Partner

📌 Presenter: Google Cloud Summit Seoul · AWS Summit Seoul · NAVER DeviewOpen to global opportunities in LLM engineering, alignment research, and AI infrastructure.

Linkedin Badge Youtube Badge Facebook Badge Gmail Badge

Pinned Loading

  1. bert_pytorch_onnx bert_pytorch_onnx Public

    Python 8 2

  2. tdd-master/tdd-refactoring-study tdd-master/tdd-refactoring-study Public

    2

  3. benchmark_keras-mxnet benchmark_keras-mxnet Public

    Python 3 1

  4. KoNLPQ KoNLPQ Public

    NLPs

    R 7

  5. pretrain-dev pretrain-dev Public

    Python 4 1

  6. KoGPT2-DINO KoGPT2-DINO Public

    20 1