Knowledgesupply chainlogistics
Supply chain & logistics foundation
Graduated in Supply Chain Management and went straight into freight forwarding. The industry runs on fragmented data and manual paperwork.
Learning Journey
From Supply Chain
to Software
Milestones, challenges, skills acquired, habits built, and knowledge gained, one year at a time.
Entry types
MilestoneChallengeSkillKnowledgeDomainHabit
2021
2022
Domaindata analysisprocess design
Discovering the systemic bottleneck
Working across sea and air operations, I noticed how much time went to things that should be faster. Started using spreadsheets seriously to pull useful info from messy data. Cut reporting time by about 60%.
2023
Skilllanguagemandarin
Attempting to learn Mandarin
Started learning Chinese. Found it really hard and eventually paused. Didn't stick, but at least I tried.
Domaincompliancetrade law
Mastering the rules of the game
Started working toward my customs broker license. Learning trade law and tariff codes taught me that compliance has to be built into the system from the start.
2024
ChallengeERPmake-vs-buy
The ERP wall and the pivot to custom build
Off-the-shelf ERPs were too expensive and too rigid. So instead of buying software, I started learning how to build it myself.
2025
MilestoneJavaScriptTypeScript
First production web apps
Went from spreadsheets to full-stack apps. Built internal dashboards the team uses daily for tracking and profit analysis.
MilestoneAIautomation
Event-driven AI automation
Set up AI to read vendor emails, thread conversations, and route inquiries automatically. Cut email sorting time by 85%.
Challengeinfrastructureself-hosted
Engineering on a budget
Hit free tier limits. Started self-hosting, picking open-source tools, and making sure nothing quietly racks up monthly bills.
Habitdisciplinework ethic
The daily grind: 12+ hours of learning
Went all in on building. Split days between operations work and coding practice. Built the discipline of shipping something every week, even when most of it got thrown away the next.
Habitcode qualitysecurity
Code review before every merge
No merge without a structured review pass. Used automated scanning and caught security gaps before they hit production.
2026
MilestoneAI agentsedge computing
Dual-agent architecture
Two AI agents running in parallel: one on a cloud server for daily ops, another on a home GPU for local inference. Connected via encrypted mesh networking.
Milestonemulti-agentcoordination
From two agents to a squad
Added a third agent on a separate workstation. Each has its own specialization: daily ops, research, and experimental work. They coordinate through shared memory instead of stepping on each other.
ChallengePythondata migration
Zero-downtime legacy migration
Built a sync engine that moves 490K records from a legacy database to the cloud. Runs hourly, fixes dirty data on the way, hasn't broken production.
Challengedatabasedata ownership
Moving to self-hosted databases
Cloud database costs add up and remote latency bothers me. Planning a full migration to self-hosted. The tricky part is doing it without breaking production.
SkillAI toolingautomation
AI delegation patterns for coding
Developed workflows where AI orchestrates subagents for parallel exploration, review, and implementation. Multi-file refactors without losing track.
SkillAI infrastructureself-hosted
Self-hosted AI gateway
Put an AI gateway on a server to stop paying per-call API fees. Powers my personal AI agent for routine tasks. Cut API costs by about 80%.
Skillorchestrationgovernance
Orchestration layer for AI agents
Multiple agents across servers needed cost tracking, task assignment, and guardrails. Building centralized orchestration -- air traffic control for automated workers.
Skilldistributed systemsautomation
Teaching agents to talk to each other
Agents on different machines communicating through shared sessions. One can spin up a task, hand it off to another server, and get results back.
Knowledgesecuritydatabase
Row-level security patterns
Learned the hard way that security policies must be on every table from day one. Audited 104 tables, found 3 without policies, and set a standard for every new table.