Open to collaboration

I design AI that
retrieves with evidence.

I'm Byoungwook Kim — AI engineer at Runbear, creator of AutoRAG (★4.9K), the open-source framework that finds the optimal RAG pipeline for your data. I work on retrieval, evaluation, and AI safety — and I write about all of it.

AutoRAG★ 4.9K on GitHub
CurrentlyAI Engineer @ Runbear
Co-founderNomaDamas hacker house
FocusRAG · Evaluation · Safety
AutoRAG repository preview

★ 4.9K2024 —

AutoRAG

RAG auto-optimization framework. Finds the best pipeline for your data — parsing, chunking, retrieval, and generation — automatically.

RAGAutoMLPythonLlamaIndex

Claude Code로 오픈소스 만들며 느낀점

[논문 리뷰] Defending Against Prompt Injection with a Few DefensiveTokens

[논문 리뷰] PoisonBench: Assessing LM Vulnerability to Poisoned Preference Data

[논문 리뷰] Cheating Automatic LLM Benchmarks: NULL Models Achieve High Win Rates

[논문 리뷰] DarkBench: benchmarking dark patterns in LLM

+25

All 30 essays

Browse the full archive.

Software Engineer · Runbear 2025 — Present Building AI teammates that live where teams work — Slack, Teams, and Discord. Connect your tools and knowledge, and an AI teammate answers, acts, and automates.
Instructor · Fast Campus — Korean DeepLearning.AI 2024.10 — 2025.03 Taught the online course "Evaluation & Optimization of RAG Systems" — how to measure and systematically improve RAG pipelines.
Co-Founder · AutoRAG Inc. 2025 Founded a startup to commercialize AutoRAG, the open-source project loved by engineers worldwide. Later pivoted from the founder role to pursue a dedicated technical path in software engineering and open source.
AI Engineer · Markr.AI 2023.06 — 2024.09 Co-led open-source projects. Designed AutoRAG's modular framework (★4.9K), engineered hybrid retrieval and parallel inference, researched chunking and LLM-as-a-Judge, ran a 500+ member Discord community, and shipped Kice-Slayer (★530) and RAGChain (★282).
Co-Founder · EDAI Inc. 2022.09 — 2023.05 Built an in-game PPL (product placement) solution as a Unity SDK. Interviewed 60 companies in 2 days, closed contracts with a metaverse company and an ad company, and filed a patent for 3D virtual-space exposure measurement.
1st Prize — 2025 SK AI SUMMIT Claude Code Builder Hackathon (Anthropic · SKT · Coxwave) Awards
1st Prize — 2024 Open-Source Developer Competition (Ministry of Science and ICT, Korea) Awards
Uber AI Online Seminar Invited Talks
AWS Gen AI Loft Seoul 2024 Invited Talks
GDG DevFest Seoul 2024 Invited Talks
Kim. D., Kim. B., et al. (2024). AutoRAG: Automated Framework for Optimization of RAG Pipeline — cited by 29 Publication

I believe the best AI is built in public. At Markr.AI I co-led the open-source projects that became AutoRAG — designing the framework, researching chunking and LLM-as-a-Judge evaluation, and growing a 500+ member community. Before that, I co-founded two startups: AutoRAG Inc., to commercialize the framework, and EDAI, an in-game advertising SDK with a granted patent.

Today I build AI teammates at Runbear and co-run NomaDamas, an open-source AI hacker house in Seoul. I teach what I learn — on stages like AWS Gen AI Loft and GDG DevFest, and in a Fast Campus course on RAG evaluation.

Now — 2026

  • Building AI teammates at Runbear
  • Researching RAG evaluation & AI safety
  • Writing paper reviews on this blog
  • Traveling as a digital nomad, when possible