writing2026-09-12

Google DeepMind Gemini3 Hackathon

writingbloghackathongemininextjsgcp

Summary

Published 2026-02-28 · slug equilaw-gemini3-hackathon-retro · category "Projects, hackathons, and competitions / Product hackathons" · source article recorded in Korean and English. Blog tags: Hackathon, Gemini API, Next.js, GCP, Docker, UX.
A retrospective on building EquiLaw at the Gemini3 DeepMind hackathon: an AI labor-law violation analysis service for workers with limited legal knowledge. Gemini analyzes multimodal evidence such as KakaoTalk screenshots and call recordings, maps it to relevant legal provisions, and generates a complaint for the Ministry of Employment and Labor with one click, including a simulation of submission by fax.
Five features: multimodal evidence analysis; an unpaid-wage calculator covering additional allowances and late-payment interest; automatic generation of a three-page complaint and PDF; office lookup, a visualization of complaint-portal entry, and a fax-sending UI; and an Agent Activity Monitor that shows logs while AI processing is underway. Stack: Next.js 15 App Router, React 19, Tailwind CSS v4; Google Gemini API (@google/genai) for text, image, and audio; Docker, GCP Cloud Run, and Cloud Build.
Troubleshooting: the Cloud Run build pipeline omitted build dependencies with npm ci --omit=dev; switching to npm ci, including dev dependencies, resolved it. PDF saving moved from html2canvas to window.print().
Result recorded in this note: deployed to GCP Cloud Run and implemented the flow from evidence analysis to legal-provision mapping, unpaid-wage calculation, complaint generation, and fax submission. The article version summarized here did not mention an award or rank. Lessons: consider native browser features first, understand devDependencies in CI/CD, prioritize UX in AI services, and explore Gemini's multimodal API. Future ideas: connect the actual complaint portal, improve prompts, optimize for mobile, and connect users with lawyers.

Mentioned

EquiLaw: the hackathon product discussed in the article.
Google DeepMind: the hackathon was called the "Gemini3 DeepMind hackathon."
Google: Gemini API, GCP Cloud Run, and Cloud Build.

Connections

Writing · Projects
Atomic notes: Removing devDependencies from CI can break the build; consider native features first · Q: What percentage of EquiLaw did I contribute, and where is its live demo?
Concepts: CI/CD · Cloud Run · Docker · Gemini API · Next.js · React · STT and speaker diarization · Tailwind CSS · i18n and multilingual support · Running a technical blog · Multimodal analysis · Legal domain (labor law and compliance) · Prompt design

Local graph

Notes 2 · Links 1

Backlinks

KRAFTON AI R&D Hackathonwriting

Source posts

20th of 1,500 Teams at the Gemini Seoul Hackathon: EquiLaw Turns a Worker's Evidence into a Complaint