Eric does things.
Human in the loop. AI in the driver's seat. Me with the emergency brake.
Hi, I'm Eric Yim. Everything here is a joint venture between me and an AI.
The workflow is simple: the AI spews fast, remarkably confident nonsense; I catch it red-handed, fix the citations it completely made up, and sign off before anything hits production.
To sum it up: It brings the typing speed. I bring the adult supervision.
If you spot a bug or a hallucination wearing a tuxedo, that's on me — the AI tried its best to gaslight me, and I fell for it.
Archive · newest first
-
Giving “Hollyoung” a face — five routes to a product IP Design notes
From a wave spirit to a dog that can load, verify, match, submit, and recover: five mascot directions, twenty-two character boards, and the deliberate reductions that turned brand stories into memorable product behavior.
-
AI interview systems — from prototype to auditable production Implementation research
A source-pinned implementation blueprint spanning CV/JD parsing, real-time voice, session state, RAG, code judging, evidence-based scoring, security, compliance, cost, capacity, and the gates from PoC to production.
-
Beisen — the minimum purchase checklist HTML report
What to leave out, which historical free benefits to ask for, and what still needs a quote. A visual minimum-purchase checklist with original sources, without hypothetical totals.
-
Xiaohongshu creator data and the closest thing to one-click publishing Feasibility brief
The official API cannot provide follower, note, or comment data, while self-run collection is possible with risk controls. For a creator material library, a Xiaohongshu mini program offers the smoothest path: bring media, copy, and topics into the publisher in one tap, then let the creator review and publish.
-
From human mentor to deliverable digital mentor Research report
A blueprint for turning authorized, field-tested mentor judgment into versioned digital mentors that can deliver continuous career services—with Skills as the execution core and data, tools, evaluation, human handoff, and governance as the shared foundation.
-
Nowcoder — how a free coding community makes its money, end-to-end HTML report
Business-model teardown of Nowcoder: a free coding-prep community of 10M+ students monetized through three B2B layers — resume sourcing (>50% of revenue), pay-per-use exam/interview SaaS, and employer-brand ads. 59 sourced references, 2025–2026 procurement records (incl. a ¥1.45M contract from a central-bank-affiliated fintech), an AI-agent pivot timeline, and honest notes on which metrics stopped updating after 2023–2024. ~2,800 lines, no tracking, no JS beyond Chart.js.
-
Mercor — $20B in 2.5 years, end-to-end HTML report
Investment-research teardown of Mercor's 5-round, 100× valuation jump from a $194M seed to a pending $20B post-money. 25 sourced references, 6 findings (capital pacing, revenue curve, supply-side scale, margin trajectory, market reputation, valuation re-rating), plus an honest limitations section and methodology. ~2,700 lines, no tracking, no JS beyond Chart.js.
-
Jack & Jill — $20M seed round, end-to-end HTML report
Investment-research teardown of Jack & Jill's Creandum-led $20M seed. 27 sourced references, 7 findings (cap stack, moat, and a side of healthy skepticism), charts, and an honest limitations section. ~2,100 lines, no tracking, no JS beyond Chart.js.