About / person

Pakon Poomson

AI Engineer

I work with AI and data, and learn best by trying to build the real thing. This workspace keeps the question, the attempt, the result, and the limitation worth remembering.

Workspace note question → build → measure → write

what I care about

Notes from the work itself

Systems that can be inspected beat systems that only sound good.
Evaluation is the interesting part: high confidence is not the same as correct.
Writing is how a build gets tested a second time.

current engineering questions

EXP 001 When an agent answers wrongly, which step of the trace actually failed?
EXP 002 If the best chunk is corrupted, does the system still answer correctly?
EXP 003 If you change the rubric, did the same system really get better?

Diagram: a user request flows through retrieval, agent, tool, and evaluation into an answer.

01 / working loop

The loop behind the notes

A visual shorthand for how the work is approached here—not a formal methodology.

01 Question Make the unclear part visible.
02 Prototype Build a path that can be inspected.
03 Measure Keep evidence and limits close to the result.
04 Write down Leave a useful trail for the next attempt.

02 / focus threads

What keeps coming up

These threads come from the topics already collected in the workspace.

01 AI Engineering Notes about designing useful, testable AI systems. 02 Data Engineering Notes about moving data through pipelines, quality gates, warehouses, and systems that people use to make decisions.

03 / current signal

Currently

Building: AI and data systems with a visible trail
Exploring: AI Engineering, data quality, and RAG
Writing: Notes from the useful and unfinished parts

04 / knowledge + evidence

Notes ask. Projects show.

The two collections stay honest and distinct: notes hold the questions, while project pages hold the build and its evidence.

Knowledge

Questions worth writing down

AI Gets Smarter Every Day. Are We Getting Smarter Too? Lessons from AI NEXT by BIZCORE
A personal account of AI NEXT BOOTCAMP by BIZCORE on 25 July 2026—and the question of whether our context, judgment, and practice are improving with AI.
How Much Electricity and Water Does One AI Prompt Use?
What happens behind an AI response — from GPUs and data centers to electricity, cooling, and the problem with assigning one number to every prompt
One Small Data Error Can Distort the Whole System
What Data Quality really means — and why it matters from spreadsheets and dashboards to ML and AI