Search Posts
Search blog posts by title, description, tags, or content.
Hyukjun Yoon
Applied AI Engineer · Product & Backend Engineer
About Me
I draw on my experience designing and operating backend systems to apply AI to products and team workflows. I have redesigned payment and authentication systems and built LLM-based diagnostics and support-assistance features, taking responsibility from design through operational improvements. I am now building collaboration between development, review, and PM agents to improve how the team handles work from request intake to deployment.
How I Build and Operate
From requirements to working features
I connect backend design and implementation with interfaces, tests, deployment, and observability. My experience operating payment, authentication, and AI services informs the scope of each change.
Workflows for agent collaboration
I define responsibilities and handoff criteria across intake, development, review, and deployment. I connect original requests, issues, PRs, and deployments, distinguishing where agents can proceed from where people need to decide.
Outcomes checked against execution evidence
I verify results with code, tests, and runtime logs. I distinguish PR creation, deployment, and actual resolution, and turn failures into reproduction and regression checks.
Experience
Day1 Company
Backend Engineer
Aug 2024 - Present
- •Designed workflows connecting request intake, development, review, verification, and deployment to reduce repeated context sharing and follow-up instructions; built development and review agents and connected them with a PM agent
- •Defined handling paths for PRD-based features and bug or support requests without a PRD, including stage ownership and conditions requiring human judgment
- •Built a development agent on the Codex harness, preserving the familiar Codex experience from daily work when interacting through Slack
- •Reused the harness's tool execution, session management, and approval workflows through Codex App Server
- •Connect Slack development requests to requirements analysis, code changes, testing, and PR creation, mapping conversations to work sessions to preserve context for follow-up requests
- •Pass the original request and recent conversation into the execution input, and assign repositories and worktrees per task
- •Built an independent review agent with Hermes Agent, connecting its Slack Gateway, dedicated GitHub and Slack MCP tools, and PR and security review skills
- •Reserved product-code modification for the development agent and designed revision, re-review, and human escalation rules based on review findings
- •Pinned reviews to a commit and checked the current code before posting, preventing duplicate reviews and blind retries when delivery is uncertain
- •Customized Gateway behavior to route progress logs separately from final replies and prevent mentions in intermediate messages from triggering other bots
- •Designed GitHub-based work management and deployment tracking
- •Implemented assignee, schedule, and progress management in a PM agent
- •Distinguished development completion from deployment, requiring deployment confirmation before closing work
- •Packaged shared skills and MCP tools for database, log, browser, and development-environment access; verified package tests and live reads through available integrations
- •Developing an E2E agent that automates per-PR test environment setup and verification of changed functionality
- •Connect verification environments combining PR branches across repositories with Playwright test selection based on changes, and design QA-specific authentication
- •Developing an audit agent to compare execution records with deliverables and identify incomplete work, omissions, and recurring errors
- •Wrote a self-improvement skill that reproduces failures and checks fixes against unchanged tests and normal cases; defined daily audit procedures to track unfinished work
EXEM
Backend Engineer
Jul 2020 - May 2024
- •Reworked MyBatis queries with subquery structures to avoid query-size limits from large parameter sets
- •Combined multiple split read queries into a single query to improve response performance
- •Implemented core business logic for a database analysis tool
- •Improved large data read performance to respond within 100ms to 3s
- •Provided database status and analysis metrics visually in the user interface