AI Toolchain: How I Built a Website and a Research Report

Confession first: I'm not a CS student

My curriculum has financial accounting, audit and financial-statement analysis — not data structures. Yet this year I shipped two things with AI tools:

• a 5-page, fully offline xianxia reading companion;
• a 7-day field research covered by provincial Party media.

The common thread: use AI to cover what I don't know well, keep judgment to myself.

My toolbox

• Scraper — raw data: novel content, enterprise websites, media reports.
• Codex — bulk cleaning: character relationships, company data.
• Claude Code + DeepSeek V4 PRO — framework building.
• Claude Code + DeepSeek V4 Flash — batch image work.
• Codex + ChatGPT — visual design (you'll see why).
• Prompt engineering — the most underrated skill.

Project 1: the reading companion

The anime updates slowly and my time is fragmented — I kept forgetting what happened earlier. So I built a tool to recall content anytime.

Pipeline: scrape full text → Codex organizes relationships & plot → Claude Code + DeepSeek PRO builds the 5-page shell → Flash batch-collects character images into cards.

Pitfall 1: DeepSeek's UI was ugly. So Codex + ChatGPT designed the "ink-bamboo green + antique gold" palette and the floating gourd animation.

Pitfall 2: BGM stopped when switching pages. Fix: rebuilt as "app-shell + content pages" architecture — one persistent player, iframe content, seamless audio. "Looks like it runs" vs "actually runs well" are different things.

Project 2: the field research

I turned research into a data pipeline:

1. Codex bulk-collects — enterprise sites and media data.
2. Preprocess + benchmark — compare with leading enterprises; gaps are the story.
3. Output pain-point & gap lists — decide what the 7-day interviews must ask.
4. Field-fill + data manual — from "0.7% extraction gain → ~10M yuan/year → 10k tons less waste" to "40 fewer workers on four packaging lines".
5. Pain × data → policy suggestions — a database and recommendation base.

Result: 21+ docs, published in Tongling Daily & Anhui Daily, selected by Xuexi Qiangguo.

The takeaway

Accounting trained me to cross-verify; AI trained me to scale. Together: framework first, reconcile, amplify with tools, and honestly admit the boundaries.