AI usage · 5 stages · Stage 3+4 in progress

Web Developer, Full-stack Developer, Data Analyst, Product Engineer, Infra Engineer, SAR / Satellite Engineer, AI-native Builder

A developer who uses AI as a thinking partner — expanding role boundaries while building systems alongside it.

I like absorbing a company's problem quickly from the sidelines and closing the whole loop — from domain learning through full-stack implementation to operations.

I cover frontend + backend + infra/CI-CD + AI end-to-end as one full-stack developer, and have extended that reach into the SAR / satellite domain I first met at work.

Main role
Web Dev · Full-stack
Learned domain
SAR · Satellite imagery
AI usage stage
Stage 3+4 in progress
Experience
2 yrs 5 mos
Where I am

Stage diagnosis starts from your main role

The same SAR work means something different when a geospatial-engineering major does it as their main role versus when a web developer does it as a domain expansion. The matrix below and the 5-stage model anchor my current coordinates.

Main role
Web Development
Frontend + backend + infra/CI-CD + AI
2 yrs 5 mos in, expanding into a generalist full-stack profile.
Academic background
Geospatial Information Engineering
Indirectly connected to SAR / satellite work (undergraduate level).
Learned at work
Sentinel · SAR · Satellite imagery · InSAR · MintPy · SNAPHU
Out-of-role areas I picked up after joining Lumir.

AI usage · 5 stages — in my own words

Current coordinates · Stage 3+4 in progress
1
Not in use / passive
Search replacement · Output owner = Human
Blocked by the company, or tried once or twice and dropped. Roughly a search replacement.
2
Delegating my own tasks
Email · summary · draft · Output owner = Human (just faster)
Cleaning up email, summarizing, drafting. Hands move faster, but the output is still 'AI generates and I polish.'
3
Expanding role boundaries
Adjacent-role work · Output owner = Human + AI
Now
How I feel it:I don't know the area perfectly, but solving one problem after another with AI grows both confidence and capability — the felt sense of stepping outside the role box.
4
Systemization
Workflow automation · Output owner = System
Now
How I feel it:Run multiple things in parallel each day, operate several AI agents at once, and even use people as agents. You move into the system-designer seat.
5
Side-project authority bypass
Bypass-via-side-system · Output owner = System + validation
Entry signal
How I feel it:Either 24/7 unattended operation, or porting side-validated patterns back into the company (the him e2e back-transfer). Beyond that, I'm sketching a paradigm shift in the interface itself — Jarvis-grade.
🔑
Single lever
do you use AI as a thinking partner?

The stages aren't sequential steps. Once you flip into thinking-partner mode — set the context → let AI ask you questions → think again → give feedback → and watch your own thinking sharpen in the loop — stages 3, 4, and 5 unlock together.

Projects

Same ingredients. Different course.

An assembled resume — the same 6 projects read differently for different audiences. Use the track toggle below to switch the depth and angle to the company / domain context.

lumir-sar-platform
Stage 3+4 integrated (3-layer full-stack)

Lumir SAR Data Platform

Unified search, storage, analysis, and request for Sentinel + LumirX — an internal satellite-data full-stack service (3 layers)

Problem

There was no integrated service to take Sentinel and LumirX data from search through storage, analysis, and request in one cycle. Storage (NAS + CDSE), InSAR analysis (SNAP · ISCE2 · MintPy), and the user-facing frontend were all separate — making it hard for a user to get from 'I want this location' to a finished result.

System

I'm solo-designing and building the 3-layer integrated service. **Storage layer** (sar-data-retrieval, NestJS monorepo + CDSE + NAS PoC + DDD 5-layer) + **Analysis layer** (lumir-linux-snap, 5-tool stack + multi-agent worktrees) + **Frontend layer** (sar-search-and-analyzer, Next.js + map + AOI + analysis-request UI). All three are 100% AI-native, and I'm back-transferring the 4-layer + CQRS + Korean-method-name pattern validated in him into the company backend design.

Impact

With ISCE2 in place, the analysis layer is fast enough that *weather- and season-independent surface-displacement data as a service* is now feasible. The key point: a user picking a location on the map gets either the stored result instantly, or a freshly-processed one — and the whole pipeline is full-stack-owned by a single person.

Layers
🗄StorageSentinel SAR search & analysis backend
AnalysisSentinel-1 InSAR processing pipeline
🖥FrontendSentinel SAR search & request frontend (full-stack-ready)
3-layer full-stackSentinel-1 + LumirXNestJS + Next.jsSpeed via ISCE2Multi-agent worktrees+3
View details
sdpe
Stage 3+4 blend

SDPE — SAR processing pipeline orchestration

Lumir LumirX multi-stage SAR pipeline · NestJS with 5 subsystems + DAG

Problem

There was no system letting an operator configure, run, trace, and recover the L0–L3 LumirX raw-SAR pipeline, and adding new satellites or algorithms needed to cost as little code change as possible. From my side, I was dropped into the pipeline domain blank-slate.

System

On top of an inherited NestJS 5-subsystem monorepo, I designed and built the DAG planning UI (no Figma — the UI code plus Playwright e2e *is* the planning document). Built the GitLab CI/CD from zero, did the detailed design for interfaces/csc-8, fed the 80–100-page ICD/SAD docx directly into AI, ran a tight loop of small 'does this break the rule?' reviews, and wired up an auto-redeploy hook that fires when ops-console work completes.

Impact

Despite entering blank-slate, I produced detailed design and implementation on top of my senior's base. The architecture has settled into a shape where adding Sentinel support or a Snappy-based DAG step takes near-zero time.

Areas
🎨
Planning · UI
DAG pipeline UI (no Figma — UI code + Playwright e2e is the planning doc)
CI/CD
GitLab CI/CD from zero (server setup + custom mail + runners)
🔌
Interfaces
Detailed design for interfaces/csc-8 (csc-7/9 to follow)
📋
Meta · consistency
ICD/SAD docx ingestion + repeated checklist-violation reviews via markdown
NestJS · 5 subsystemsDAG pipelinepgmqGitLab CI/CDICD/SAD consistency mechanism+1
View details
him
Stage 4 + 5 (back-transferred into work)

him (home-inventory-manager)

100% AI-native full-stack side project

Problem

Real household-inventory pain points (where did it go · expiration dates · stock levels · re-order alerts), plus a personal motive: prove I can solo every layer — UI planning, frontend, backend, infra.

System

NestJS + CQRS + TypeORM + PostgreSQL + a frontend + Docker Compose + S3 + Terraform IaC + backup metrics → Grafana. Architecture patterns are lifted directly from work-validated ones. 100% AI-native: AI writes code; the three human nodes are (1) UI planning, (2) code-connection review, and (3) e2e tests. Building self-hosted infra so DB-level user isolation works.

Impact

Personal use + 1 beta user (a shared hub), 13+ wiki assets fed back (the 2-stage design process, the 5-layer testing setup, 31 consistency-gap repairs, automated backup metrics, and recipes for both sides of S3 + Terraform). Back-transfer into work happened too — the him UI + current e2e test pattern is being adopted to strengthen the company frontend, and the 4-layer pattern is being applied to the sar-search-and-analyzer backend design.

Areas
🔧
Backend
NestJS + CQRS + TypeORM + PostgreSQL (patterns lifted from work)
🖥
Frontend · e2e
UI + current e2e tests (being back-transferred into the work frontend)
Infra
Docker Compose + S3 + Terraform IaC + Grafana backup metrics
📚
Wiki feedback loop
13+ pages fed back + 31 consistency gaps repaired
100% AI nativeNestJS + CQRSTypeORMTerraform IaCAutomated backup metrics+3
View details
lumir-erp
Stage 4 (in-role · adapting to 4 domains)

Lumir-ERP (internal back office)

Four domain workstreams together — full-stack CMS + frontend for resource scheduling, LRIM, and interview management

Problem

We needed full-stack delivery of four internal back-office domains (resource scheduling · CMS · LRIM hiring · LRIM interview management). It's the company's broadest-user project — every employee, plus recruiters, interviewers, evaluators, and external applicants.

System

Next.js 14/15 (App Router) + TypeScript + Tailwind + shadcn/ui + SWR + Playwright. A Plan (mock) / Current (real API) environment-split pattern (inherited from the team lead) + per-domain Context + (cms)/(sms)/(ams)/(uam) domain separation + UAM uses MongoDB directly. The two LRIM apps form a pnpm + Turborepo monorepo (@repo/ui · common · modules). 100% AI-native: AI writes code; I review the UI planning, code connections, and own the e2e tests.

Impact

A 4-domain back office running in production for every employee plus external applicants. CMS is solo full-stack (planning · backend · tests; frontend tests are being improved). For the other three, I own all frontend functionality. The real asset isn't 'four projects' — it's the ability to adapt to four domains in parallel and the full-cycle solo experience on CMS.

Areas
🗓
Scheduling
calendar / schedule-status — all frontend (for every employee)
📝
CMS
Solo full-stack — planning · backend · tests (frontend tests being improved)
👥
LRIM hiring
All frontend, for recruiters · interviewers · evaluators
📋
LRIM interview mgmt
Applicant-facing interview-slot selection (publicly exposed)
Next.js App RouterPlan/Current environment splitSolo full-stack CMSLRIM hiring & interviewsExternal applicants+3
View details
brain-trinity
Stage 4 + stage-5 signal

Brain Trinity

Systematizing cognitive load distribution + a domain-adaptation mechanism

Robotics track only
Problem

To keep my biological brain on the important work, the 'important but doesn't need to live in working memory' content had to be offloaded to an external system. Plain note apps are searchable but don't synthesize, link, or re-use (trigger: Karpathy's LLM Wiki video).

System

Karpathy-style LLM Wiki 3-layer (immutable raw/ + AI-compiled wiki/ + derived Output/) + Claude Code as the only collaborator + a skill system (ingest/lint/query) + a frontmatter schema + auto-updating index/log + Obsidian graph + auto-syncing MEMORY. 100% AI-written; the user's input is just prompt chat.

Impact

56+ wiki pages and counting; every ingest triggers automatic cross-linking, raw-frontmatter sync, and index/log updates. This resume, the self-diagnosis, and the 6 project briefs are themselves the living evidence of Brain Trinity's stage-4 automatic operation. Bundling internal projects (sdpe + lumir-sar-platform) on top would open a stage-5 in-house systemization flow.

Areas
📥
raw/ layer
Immutable raw inputs dropped by the user (chats · papers · videos · notes)
🧠
wiki/ compile
AI compiles raw into wiki pages (cross-link + index/log automated)
Skill system
ingest / lint / query — explicit workflow automation
🔄
MEMORY sync
Context preserved across sessions via Claude Code MEMORY
Cognitive load distributionKarpathy /raw patternClaude Code onlySkill systemAuto-syncing MEMORY+1
View details
Now showing · 5 projectsTotal · 5 (Brain Trinity shows only on the robotics track)
Tech stack

Main role · Learned domain · Future

I separate my main role (web development) from the learned domain (SAR · satellite imagery) picked up at work. Depth-within-role and role-expansion are two distinct things.

Main role · Web Development

Stage 4 · systemization
Frontend
  • TypeScript
  • Next.js
  • React
  • Tailwind CSS
  • Playwright (e2e)
Backend
  • NestJS
  • CQRS· @nestjs/cqrs
  • TypeORM
  • PostgreSQL
  • pgmq· PostgreSQL message queue
  • DDD 5-layer· domain / business / context / handlers / interfaces
  • Jest + Testcontainers
Infra · CI/CD
  • Docker · docker-compose
  • GitLab CI/CD· Built from scratch + custom mail
  • Terraform IaC· Recipes for both S3 sides
  • Grafana + Prometheus· Backup metrics via textfile collector
  • FastAPI· Analysis server (bridged to NestJS)
AI collaboration
  • Claude Code· Primary thinking partner · 100% AI native
  • 다중 agent 워크트리· Agents 1–4 in parallel + handoff system
  • Brain Trinity 위키· Karpathy /raw pattern + skill system

Learned domain · SAR · Satellite imagery

Stage 3 · role expansion
SAR · Satellite imagery
  • Sentinel-1 SAR
  • ESA SNAP 12· MicrowaveTBX (SAR)
  • SNAPHU· Phase unwrapping
  • MintPy· SBAS time series
  • ISCE2· New track — speed unlocked
  • StaMPS PSI· Octave + 12 patches
  • DInSAR · SBAS · PSInSAR
  • PyAPS + ERA5· Atmospheric correction
  • CDSE· Copernicus Data Space
Domain tools
  • QGIS
  • Python (분석)· rasterio · geopandas · shapely
  • Snappy· SNAP Python bridge
Next role expansion — projected
Embedded (planned)Hardware (natural extension)Robotics (industry-transition target)
About

Changuk Woo

Web Developer, expanding into SAR. I cover frontend + backend + infra/CI-CD + AI end-to-end as one full-stack developer, and have extended that reach into the SAR / satellite domain I first met at work.

Are you launching into a new industry that demands fast domain absorption?

Do you need a full-stack engineer who treats AI as a collaborator, not just a tool?

Looking for someone who owns one full cycle — code, planning, infra, and testing?

Brain Trinity · live demo available

This portfolio, the self-diagnosis, and all 6 project briefs were compiled from a meta system called Brain Trinity. It uses the Karpathy LLM Wiki pattern + Claude Code collaboration + a skill system, and currently holds 56+ accumulated wiki pages.

Given an unfamiliar domain, I believe structured information + an AI-native foundation can solve almost anything. Brain Trinity is both the methodology and the asset — and I can demo it live in an interview.

What's next
  • Currently extending into embedded systems
  • Expecting a natural extension into hardware and robotics
  • Plan to grow Brain Trinity into a complete personal system (voice + journal + meeting notes + PDFs unified)