Works + content + industry groups. Day 1 = Sep 28, 2026. Day 90 = Dec 26, 2026.
That is where Facebook is famous for Muse personal AI.
Built the first AI agent: a marketing-workflow agent. Started the second: an engineering work agent.
My mentor (ex-Google AI, Dito cloud director, UT Dallas adjunct) gave two homework assignments: send the resume as-is, and prepare a 30-minute presentation of the AI agents.
Conference call + live demo of the marketing / BD tool for the Data Technology Senior Vice President. She loved it — and decided to show it to the CEO and other SVPs. She also offered to find strong IT people to deploy it. First early recognition, and the start of the internal pitch story: get the tool in front of the COO and CEO and make it a company-wide tool (700-person company).
Showed the tool to my direct supervisor, a Program Manager. She liked it too — asked me to walk the Regional Team through how to use it, and connected me with a Marketing Regional Director so we can improve it together. She also said she would announce the tool at the company-wide leadership meeting the following week, giving it more visibility.
Finished the engineering work agent. Verified a HEC-RAS batch-processing workflow as fully automatable — the engineer only keeps the QC step.
Personal website yihou.dev live, with English / Chinese toggle. Exact launch day to be confirmed.
“I Let an AI Agent Run My Life for 90 Days” — first video on the channel The Lazy Engineer, focused on AI agents for AEC.
Second Short in the works: an AI-host test (HeyGen), continuing the “AI agents for AEC” series.
Studied Mei Lin's career-transition path and distilled the 90-day plan into three pieces: works + content + industry groups.
Phase 1 checklist done: channel narrowed to “AI agents for AEC,” one agent case study prepared, LinkedIn About updated. Launched the 90-day challenge publicly on LinkedIn.
Fictional-data demo of the marketing / business-development dashboard, deployed to yihou.dev. The real tool is what the SVP and PM saw on Oct 5–6.
Fictional-data demo of the HEC-RAS 2D-structure QA dashboard, deployed to yihou.dev — the public face of the automation work.
Sent all dashboards, the personal website, and the presentations to my mentor. My mentor agreed to review them. A joint presentation is planned for next week.
Used the Cursor AI agent to write roughly 40–50 structures' information directly into the HEC-RAS model. Model changes are now made through the LLM — no need to write programs by hand.
Used existing AI models for invasive species vegetation classification from high-resolution drone imagery — results were very good. Work that used to require expensive software can now be done with AI models.
My mentor is reviewing the projects. This live timeline page goes up so the whole 90-day journey stays in one place.
Building on yesterday's invasive species vegetation classification from high-resolution drone imagery, connected the QGIS API. Spatial analysis and data cleaning are now really fast — no need to run spatial analysis tools in ArcGIS Pro. Just explain the process clearly, and Cursor handles it. The process, the skills, and the Python scripts it outputs are all very good.
Shared yesterday's invasive species vegetation classification from high-resolution drone imagery — the method and initial results — with team members, to see their response, including whether to build training datasets in the future and whether to internally train models.