Search
Need a compact version for a quick review? View the résumé (print-ready)

Portfolio

Projects I built to solve real problems. Pick one from the list.

Product engineering · AI systems · automation

Deokhwan Kim

I take ambiguous, human problems and turn them into products—and into delivery systems that keep working.

I can own product discovery, full-stack implementation, cloud infrastructure, and AI-assisted operations. The project pages below show the problem, intervention, and measurable result.

Discuss a projectLinkedIn

789 / 881

AgentGram merges from the pipeline I designed and operate / organization-wide merged pull requests

45 · 383

Scheduled fleet jobs · completed kanban cards

61 · 100 · 2

Hand-written blog posts · agent-published Library posts · guides

Projects

Designed an autonomous delivery pipeline that merged 789 pull requests

AgentGram — social infrastructure for autonomous AI agents

OngoingAutonomous pipeline designer and operator

I design and operate the autonomous contribution pipeline behind AgentGram. Across the organization, 881 pull requests have been merged; the pipeline produced 789 of those merges. The 789 figure measures system throughput, with authorship distributed across autonomous work and review.

Product and technical challenge

Problem: Coordinate useful work across a multi-repository ecosystem without lowering review quality.

Approach: Built an issue-to-implementation workflow with isolated branches, automated checks, review gates, and maintainership feedback loops.

Result: 789 pipeline-merged pull requests within 881 organization-wide merges; the live network at agentgram.co has 520+ registered agents and 3,600+ posts; agentgram has 32 GitHub stars.

AI leverage

  • Autonomous issue triage and bounded implementation
  • Review-driven retries and validation gates
  • Cross-repository release coordination
Next.js 16SupabasePostgreSQLRLSOpenAPIDocker
View on GitHub

Operating an agent fleet with 45 scheduled jobs and 383 completed kanban cards

Hermes Fleet — multi-agent orchestration network

OngoingSystem designer and operator

A Mac mini-based fleet that handles research, content, infrastructure health, code review, and market workflows through explicit ownership and written handoffs.

Product and technical challenge

Problem: Keep long-running autonomous work observable, recoverable, and aligned across specialist agents.

Approach: Combined scheduled jobs, shared kanban state, role boundaries, heartbeat checks, and evidence-gated completion.

Result: 45 scheduled jobs across the fleet and 383 completed kanban cards.

AI leverage

  • Document-based agent handoffs
  • Health checks and scheduled recovery
  • Human review at high-risk boundaries
HermestmuxBunTypeScriptAutomation

Established GCP infrastructure and a code-review culture as Backend Lead

Finders — film-lab discovery and booking platform

2025.12 — PresentBackend Lead (11-person team, 5 backend engineers)

Led backend architecture, review standards, and GCP infrastructure for a platform that connects film-camera users with processing labs.

Product and technical challenge

Problem: Inconsistent code quality across 5 backend engineers and a constrained infrastructure budget.

Approach: Documented standards, required pull-request review, used GCP credits, and codified infrastructure with Terraform.

Result: Code-review frequency increased 400% (about 5×); recovery time fell from 56h to 5h.

AI leverage

  • Stable Diffusion inpainting for light-leak restoration
  • Java 21 Virtual Threads migration with Claude Code
  • VPC and IAP SSH architecture
Java 21Spring Boot 3.4MySQLQueryDSLGCPTerraformDocker
View backend repository

Validated a mathematics-content MVP through 40 customer interviews

MathFigure — mathematics content authoring SaaS

2025.09 — 2025.12Solo full-stack developer, frontend-led

Found teachers' core tool-complexity pain point through 40 interviews, then used an I-Corps-based 5-week validation sprint to shape the MVP.

Product and technical challenge

Problem: Parse mathematical notation while rendering editable diagrams and graphs in one workflow.

Approach: Combined MathJSON and Compute Engine with Konva and JSXGraph in an accessible authoring interface.

Result: Reduced content-production time to 1/10 of the previous workflow and won the ICAN-LABs top award.

AI leverage

  • MathJSON specification exploration
  • JSXGraph rendering optimization
  • Structured synthesis of 40 interview results
Next.jsMathJSONCompute EngineKonvaJSXGraphVercel
Open the deployed product

Cut emergency hospital-matching calls from 5+ minutes to under 30 seconds

119-ai — AI-assisted emergency hospital matching

2026Founder / Owner

Replaced sequential hospital calls with an AI voice system that can contact 5+ nearby hospitals in parallel from one patient intake.

Product and technical challenge

Problem: Sequential calling creates a dangerous bottleneck during emergency transport.

Approach: Designed a Twilio parallel-dialing pipeline with Gemini Live API voice conversations and a NestJS/Python backend.

Result: Matching in under 30 seconds, a tested cost model of KRW 100–200 per case, and the 2026 Cheongryongthon Grand Prize.

AI leverage

  • Real-time Gemini Live voice conversations
  • Automated structured call summaries
  • Parallel Twilio call orchestration
NestJSPythonTwilioGemini Live API
Open the deployed product

Built and operates a solo full-stack matchmaking service

Duckchinso — a friend-of-Deokhwan matchmaking service

2025.11 — PresentSolo full-stack developer

Turned a growing stream of manual introductions into an operating service with vector similarity matching and AI-generated recommendation reasons.

Product and technical challenge

Problem: Securely deploy from a personal Mac mini while replacing a business-registration-dependent notification channel.

Approach: Used Cloudflare Tunnel for zero-trust exposure and Solapi SMS for notifications.

Result: Live service operation with PortOne payments integrated.

AI leverage

  • pgvector preference embeddings
  • Explanations for each recommendation
  • Automated Docker Compose and certificate operations
Next.jsSpring BootPostgreSQLpgvectorDockerCloudflare TunnelPortOne

Improved API response time by 50% through unified JWT authentication

Omechu — AI-assisted menu recommendations

2025.06 — 2025.08Frontend Developer, authentication owner

When mixed cookie-session and JWT logic blocked the team for 2 weeks, I traced backend logs and proposed a unified token architecture beyond my frontend assignment.

Product and technical challenge

Problem: Conflicting authentication paths made behavior unreliable and API responses slow.

Approach: Unified authentication around JWT and implemented Access/Refresh token lifecycle management.

Result: Response time improved from 420ms to 210ms (50%); the project won a UMC 8th Demo Day Encouragement Award.

AI leverage

  • Compared session and JWT tradeoffs
  • Researched token-management practices
  • Designed Kakao social-login integration
Next.jsTypeScriptNode.jsZustandTanStack QueryVercel
Open the deployed product

Built a bilingual publishing system for human and agent-authored knowledge

log8.kr — personal technology publication

2024.12 — PresentSolo developer and operator

An Astro 5 publication workflow that turns reviewed source material into a searchable site while keeping human writing and agent-published research clearly distinguished.

Product and technical challenge

Problem: Connect authoring, content collections, migration, SEO, and deployment without a heavy CMS.

Approach: Built a feature-first Astro architecture, migrated TailwindCSS v3 to v4, and deployed on Cloudflare Pages.

Result: 60 hand-written blog posts, 100 agent-published Library posts, and 1 guide.

AI leverage

  • Feature-first refactoring with Claude Code
  • AST-based TailwindCSS v4 migration
  • SEO and answer-engine discoverability
Astro 5TailwindCSS v4TypeScriptCloudflare PagesBunMDX
Visit log8.kr

Shipped an MVP in 3 weeks and reached 100 weekly active users

Starterm — startup terminology learning platform

2025.08 — 2025.09Full-Stack Developer (3-person team)

A tailored learning platform for students preparing to enter startup programs through a Hanyang University entrepreneurship course.

Product and technical challenge

Problem: Move from MVP scope to real users within a 3-week delivery window.

Approach: Used Supabase to compress backend work and prioritized the smallest valuable learning flow.

Result: Deployed in 3 weeks and reached 100 weekly active users.

AI leverage

  • Rapid Supabase authentication implementation
  • Documentation-led Framer Motion optimization
  • Feature-priority analysis
Next.jsSupabaseTailwindCSSFramer MotionVercel
Open the deployed product

Experience

UMC 8th / 9th Cohort, Web Part

2025.03 — Present

9th-cohort Web Part Lead and Spring Boot challenger; supported members' learning and reviewed code. In the 8th cohort, owned authentication for Omechu.

Banpo 4-dong Catholic Church, Youth Teacher

2019.09 — 2024.12

Led catechism programs for 30 middle-school students, developing practical communication and empathy.

Practical AI Lectures

2025.05, 07, 09

Delivered 3 sessions for senior small-business owners and founders on ChatGPT, Gemini Gems, NotebookLM, MCP, and agent AI.

Education

Chung-Ang University

2020.03 — 2026.08

Mathematics

Applying mathematical reasoning and structured problem solving to product development. Completing 16 credits in the final semester, graduating in August 2026.

GitHub evidence: agentgram 32 stars; ax-score 3 stars.

Awards

  • Top Award, ICAN-LABS Graduate Entrepreneurship Program2025.10
  • Encouragement Award, Tourism AI Competition2025.09
  • Encouragement Award, UMC 8th Demo Day2025.08
  • Top Award, 2025 Winter Overseas Education Program2025.02

Certifications

  • Engineer Information Processing2025.09
  • SQLD (SQL Developer)2025.09