Medium

Ted Park

Ted Park

·

2d ago

A Restarted Worker Must Not Submit the Same Experiment Twice

The recovery decision belongs to the retained evidence

AGENTS / RECOVERY: A worker restarts. An intent survives. Diagram: RETAIN INTENT → READ STATUS → RESUME OR WAIT.

Ted Park

·

Nov 6

A Private Local Model Endpoint Is More Than an HTTPS Certificate

How I added a VPN-only HTTPS front door to my ASUS P14 inference setup — and what the checks actually established.

LOCAL AI / PRIVATE HTTPS: A private endpoint. A verified path. Diagram: TAILNET → HTTPS PROXY → LOCAL MODEL.

Ted Park

·

Nov 6

The Local LLM Setup I Settled On: A DGX Spark Recipe on My ASUS P14

Running Qwen3.8 Flash Next on Omarchy became a question of memory, recovery, and a configuration I could understand.

LOCAL AI / PART TWO: The setup I settled on. ASUS P14 → SPARK RECIPE → LOCAL CODING.

Ted Park

·

Nov 6

Reviewing an AI-Assisted Change: What the Evidence Actually Proves

A green check answers a specific question

AI / CODE REVIEW: What does the evidence prove? Diagram: CLAIM → CHECK → LIMIT.

Ted Park

·

Nov 3

Turning My ASUS ProArt P14 into a Local AI Lab

Installing Omarchy, running Qwen3.8 Flash Next FP4, and connecting coding agents on a 128GB ARM laptop.

A laptop-sized AI lab: Omarchy boot and recovery, local inference, and coding agents on an ASUS ProArt P14.

Ted Park

·

Nov 3

A Code-Generating Agent Needs to Remember Why Its Last Attempt Failed

Separate rejected code, duplicate proposals and evaluated failures before choosing the next experiment.

CODE GENERATION: Remember the failure. Choose what comes next. Diagram: PROPOSE → RECORD OUTCOME → REVISE OR STOP.

Ted Park

·

Oct 27

A RAG Failure Needs a Location Before It Needs a Better Prompt

One bad answer, several different causes

RAG evaluation: Find the failure. Then fix the prompt. Diagram shows retrieval, context, and answer as separate stages.

Ted Park

·

Oct 15

Building a Reflective Research Agent with Elixir, OTP and Oban

A useful improvement loop remembers why it continued — and knows when to stop.

ELIXIR / REFLECTION: Reflect. Improve. Know when to stop. Diagram: EXPERIMENT → REVIEW EVIDENCE → BRANCH OR STOP.

Ted Park

·

Oct 15

Book Writer: Keep the Original When an AI Revision Loses Structure

A fluent revision can still damage a chapter

BOOK WRITER: Protect the source. Review the revision. Diagram: AI REVISION → STRUCTURE CHECK → ACCEPT OR KEEP.

Ted Park

·

Oct 13

Agent Memory and Recovery: What Must Survive a Worker

Why I keep conversational context, execution evidence, and workflow progress separate.

MEMORY / RECOVERY: What survives a worker? Diagram: CONTEXT → EVIDENCE → RUNTIME VERSION.

Ted Park

Ted Park

Senior software engineer building AI agents, RAG and backend systems. 15+ years shipping products. Writing about implementation, recovery and AI-assisted work.

Following

Help

Status

About

Careers

Press

Blog

Store

Privacy

Rules

Terms

Text to speech