[Part 1] How I Built an AI-Powered Personal Asset Dashboard

 Building a One-Person AI Content Studio for $20/Month

I built a personal asset-management system to reduce the spreadsheet work I repeated after the Korean and U.S. markets closed. The result was a web dashboard for the overall picture and a Telegram summary after market close.

Data that could be updated reliably was automated. Cash-like assets that were difficult to connect through APIs remained manual. The objective was not to maximize the automation percentage. It was to create a workflow I could actually keep using.

I did not type the program code myself, but that does not mean no code was involved. I defined the problem, the review criteria, and the disclosure boundaries. ChatGPT organized the requirements. Codex created and revised the code and files. I then used the system and decided what needed to change.

This article shows selected holdings, rounded asset values, allocation percentages, and principle-based AI review messages. It does not disclose account information, exact balances, private trading triggers, target prices, URLs, IP addresses, ports, or tokens.

The system is a personal monitoring tool. It does not provide investment advice, execute trades, guarantee returns, or recommend financial products.

The repetitive spreadsheet problem

My previous process was to review each account, enter the numbers again in a spreadsheet, calculate totals and cash allocation, and compare them with the previous day.

Existing brokerage applications are enough if the goal is only to check balance and return. I wanted a combined view of Korean and U.S. positions, cash, concentration, and a concise after-close summary based on my own review rules.

Before and after

Before the system, I manually entered data, reviewed accounts, calculated cash ratios, and repeated the same checks.

Afterward, I could review the overall flow in a web dashboard and receive a summary through Telegram. Assets that could not be connected reliably were added manually, and issues discovered in daily use were recorded for the next revision.

The first Telegram report succeeded on June 30, 2026. That was the moment part of the daily review process became operational rather than theoretical.

The human–ChatGPT–Codex workflow

• Human / CEO: define the problem, concept, priorities, facts, quality bar, and final approval
• ChatGPT: structure requirements, design the explanation, review disclosure risks, and organize next actions
• Codex: execute code, files, documents, packaging, and change records

The collaboration model used to build the personal asset-management system.

This is also a form of vibe coding, but the important part is not simply giving natural-language instructions. A useful system still needs a clearly defined problem, explicit boundaries, and repeated real-world review.

Organize before automating

1. numbers I checked every day;
2. data that could be fetched consistently;
3. information that should remain manual;
4. the rules used to review the result.

That structure made it possible to start small and expand based on actual use. For low-frequency or difficult data, manual input was more stable and less expensive than a forced API connection.

How the system works

The web dashboard shows the overall asset flow, cash allocation, and concentration in selected holdings. Connected data updates on schedule. Information that cannot be connected reliably is completed through a separate input screen.

After market close, the system creates a Telegram summary. I can review the scale, cash allocation, and items that need attention without opening the full dashboard every time.

A public version of the real dashboard. Selected holdings and rounded summary values remain visible; personal information and strategy-critical numbers are protected.

A public version of the Telegram report received on June 30, 2026. Identifiers and detailed trading criteria are protected.

The dashboard and Telegram images were captured on different dates, so their rounded totals do not match. That is a difference in data date, not a calculation error.

Real use mattered more than the first screen

Once I started using the system, I found connection issues, period-comparison needs, allocation questions, and data that was still difficult to automate. Those problems were clearer in daily use than in the original design.

I recorded each issue, revised the system, and used the new result again. A small working system improved through use was more valuable than a large design that had never operated.

What remained manual

Not every data point should be automated. Cash-like assets with limited API access or low update frequency remained manual.

A recreated manual input screen using non-sensitive sample data.

The right question is not “How much did I automate?” It is “Did the workflow reduce repetitive review without losing important information?”

A checklist for starting small

• List the numbers you check repeatedly.
• Identify what you re-enter in spreadsheets.
• Separate data that can be fetched from data that should stay manual.
• Choose a dashboard or message channel for the result.
• Keep the first release small.
• Use it, record problems, and turn those problems into the next revision.

From asset management to content publishing

The same operating model can be applied to content. In Part 2, I use AI-agent collaboration to research, script, assemble, publish, and measure YouTube Shorts.

The important result of Part 1 was not total automation. It was turning a repeated spreadsheet task into a system that could be used, reviewed, and improved.

Previous: Prologue — How I Built a $20/Month AI Content Creation Workflow

Continue: Part 2 — What I Learned Publishing Three AI YouTube Shorts

This article is an English localization of a production record first published in Korean. Korean original / 한국어 원문.