Building Chaewoon
with AI

01

From input to deployment checks

I built the path from input to a reading with the goal of clearly explaining temperament and the current life phase. I owned this personal product’s planning, brand, interfaces, backend and deployment, using AI for detailed design and implementation.

1. 01

   User input

   The user provides input that feeds the calculations for the reading.

2. 02

   Server calculations

   The server calculates Five Elements results. The generation model uses those results as the factual data for its explanation.

3. 03

   AI explanation

   The generation model writes the explanation. Automated checks verify required facts and output format; a separate AI reviews it against the sources and assesses readability.

4. 04

   Results screen

   The calculated results and explanation are presented for the user to read. The records document checks of the results screen on desktop and mobile.

5. 05

   Deployment checks

   After deployment of the web and server changes, files were compared with actual responses and screens were checked for display and behavior.

My goal and criteria

I set the goal of clearly explaining temperament and the current life phase. I adopted AI’s proposals to state the conclusion before the reason and to separate writing and review into different contexts.

AI’s contribution

AI turned the goal into writing principles and review criteria, then designed and implemented the generation, checking and review process.

Completion checks

Reviews of generated samples for required facts, format and readability were connected to post-deployment file, response and screen checks. The execution records below date from August 24–25, 2026.

At the time, an AI reviewer in a separate context compared 12 generated samples with the sources for factual consistency and readability. The execution records document deployment of the web and server changes after the samples passed AI review, followed by comparisons of deployed files with actual responses. These records cover generated samples, not user testing with 12 customers.

The same records document checks of the display and behavior of the home, results, history, payment-status and guide screens on desktop and mobile.

Execution records from August 25, 2026
runtime-r8/gate-r8.md · live-verification.md
Web 7ac02970 · Server 0befed9

02

Criteria for writing and reviewing explanations

The feature needed to return both calculated facts and readable explanations. I required automated-check results and readability to be assessed separately.

1. Turn the explanation goal into writing principles

AI turned the goal of explaining directly into a principle of stating the conclusion first and giving the reason in the next sentence. I adopted that principle. The AI-written voice guide calls for making the actor, action and object explicit and giving each sentence one job.

2. Separate writing and review into different contexts

In the process AI proposed and I adopted, the generation model writes from the server’s Five Elements results and the current writing criteria. Automated checks verify required facts and output format. An AI reviewer in a separate context compares the explanation with its sources for factual consistency and readability. AI handled the detailed design and implementation of this process.

03

Knowledge and delegation used to build the feature

To keep later changes consistent, AI needed to find both the facts calculated by the server and the writing criteria for explanations. I defined the scope and completion criteria for reorganizing the documents and assigned the work to AI. The documents were organized to point to the sources needed for writing, implementation and review.

Brand and domain knowledge

Connects source material for readings, writing principles and approved examples. It distinguishes human decisions from AI interpretations and labels assumptions about customers as hypotheses.

Task instructions

Defines AI responsibilities, required material, deliverables and completion criteria for research, copy, interfaces, server work and review.

Web and server sources

Links to the code and documents that govern actual behavior. I kept brand documents from redefining facts established by the server.

Work and review records

Preserves document changes, AI reviews and deployment checks as historical records. Current criteria and the history of past work can be found separately.

In this structure, the generation model receives current facts, approved examples and the desired result. The reviewer receives the sources for comparison and returns the conditions that a revision must meet.

AI work reports and commits record the reorganization of the specified documents, labeling of customer hypotheses and completion of link checks. Writing principles and approved examples serve as current criteria, while server-established facts and historical records lead back to their respective sources.

04

What the records establish