Backend engineer — reliable, secure infrastructure for AI systems
distributed reliability · data consistency · runtime correctness, security & cost accuracy for LLM / agent platforms

🌍 Based in South Korea — open to global roles (remote or relocation)

· github.com/Acacian/aegis

Solo-designed Python library that enforces guardrails, a policy engine, and audit logging on every LLM call and tool execution — at the CI/CD layer. Auto-instruments 12 AI frameworks (LangChain, CrewAI, OpenAI, Anthropic, LiteLLM, Google ADK …).

flowchart LR
  C["LLM call / tool exec"] --> AE["Aegis"]
  AE --> GR["Guardrails"]
  AE --> PO["Policy engine"]
  AE --> AU["Audit log"]
  AE --> D{"allow / block"}

_{Integrates as a third-party guardrail provider for Pydantic AI and LiteLLM · listed in awesome-mcp-servers and Awesome-LM-SSP.}

Fixing real correctness & cost bugs in the tools I use in production:

• litellm — bill xAI from its reported cost instead of stale recomputation (#36281)
• Spring AI — DeepSeek tool-call content=null fallback (#3817) · RedisVectorStore BuilderCustomizer (#3809)

In production today — backend for an enterprise LLM gateway / AI-security platform (Soosan): proxy request lifecycle, DLP and guardrail enforcement, cross-provider correctness, policy & audit paths, load and reliability testing.

CandyPod — backend engineer on a mobile matching app, live on the App Store and Google Play.

distributed reliability · fault isolation & recovery · data consistency · LLM / agent runtime security · cross-provider correctness & cost accuracy · AI-native system design

• Soosan (current)
• Koosstech — Backend & DevOps (E2E ownership)
• MementoAI — Backend (intern)
• ROK Air Force — mission-critical systems operations

_{Earlier: PassionPay — MSA fintech payments platform (lead), payment-service architecture & consistency design.}

Tech blog → https://victorica.tistory.com/  ·  📧 koo9811@naver.com  ·  💼 linkedin.com/in/otkling