From Jury Services to AI Builder in 6 Months

#devchallenge #googleaichallenge #portfolio #gemini

This is a submission for the New Year, New You Portfolio Challenge Presented by Google AI

About Me

I'm a Judicial Services Manager at Santa Barbara Superior Court. Six months ago, I started building AI projects. I've shipped seven, won two hackathons, and learned that the fastest way to learn is to build things that solve actual problems.

The first one started from frustration. People called constantly with jury eligibility questions. "Am I eligible?" "What does this code section mean?" "Can I defer?" I'd look up the same answers in California legal code and send emails. Same questions, hundreds of times.

So I built a chatbot. AWS Bedrock, Claude AI, fed it the legal code as context so it could answer accurately and cite sources. Took a weekend. The court staff use it now.

That project proved something: I could build useful things. From there, each project solved something I saw.

This portfolio documents six months of that.

Portfolio

Everything is live and interactive. Click through the projects, expand the side quests, try the case study generator.

The 4 Main Projects (timeline view):

Jury Eligibility Chatbot (July) - RAG-powered, uses court-verified legal code, production-ready
PDF Text Extractor (August) - Cal Poly DxHub project. I was selected 1 of 100 from 1,300 applicants. Our team placed 2nd out of 14.
Janus-Clew (December) - Won AWS Vibe Hackathon. Built a growth tracker that measures technical complexity progression.
Athena-Clew (January) - Gemini 3 Hackathon entry. Autonomous debugging tool using Deep Thinking models.

Plus 10 Side Projects: Court innovation tools, hackathon submissions, creative builds. Collapsible in the portfolio so it doesn't overwhelm the narrative.

How I Built It

Tech Stack:

React 19 + TypeScript (frontend)
Node.js + Express (backend)
Gemini 2.0 Flash API (case study generation)
GitHub API (live project metadata)
Docker + Google Cloud Run (deployment)

The Feature I Built For Myself: Dual-Mode Case Study Generator

When you finish a project, you need to explain it in completely different ways depending on your audience.

For a job application, you write:
"Led development of PDF extraction engine using Python and PyPDF2. Achieved 95% accuracy on academic document extraction. Selected as 1 of 100 fellows from 1,300 applicants. Delivered 2nd place finish on high-performing team."

For a personal blog, the same project becomes:
"My teammate Cai and I had 36 hours and an insane goal. We stayed up until 1am debugging a .env file that made no sense. But Friday morning it worked. Our team placed 2nd. Here's what we learned."

Same project. Completely different narratives. I was writing both versions and it was tedious.

So I built a tool: input your project details once, then generate either mode instantly.

Blog Post Mode produces a personal narrative (your thinking, decisions, learnings). Good for LinkedIn or your blog.

Professional Mode produces a professional case study (problem → solution → technical approach → results). Good for resumes and applications.

It's a prototype showing how AI can solve the context-switching problem developers face when marketing their work. Powered by Gemini 2.0 Flash because it's fast (under 2 seconds) and produces markdown I can actually use.

Other Design Decisions

Why Gemini 2.0 Flash? Speed matters for interactive UX. Quality output means I'm not editing markdown for 20 minutes after generation. The free tier covers a portfolio demo.

Why GitHub API integration? Live data (stars, language, forks) instead of static snapshots. Makes the portfolio feel current, not frozen.

Why Cloud Run? Meets the challenge requirement, free tier works great, and the embed feature makes it trivial to show working code.

Why this design language? Editorial style (not corporate) with intentional color psychology. Tech blue for foundation, innovation purple for sophistication, achievement gold for wins, energy cyan for interactivity. Every choice serves a purpose.

How fast? Built in ~12 hours using Antigravity (Google's AI development environment with a Chrome extension for real-time browser testing). The extension was the force multiplier—caught bugs instantly instead of the typical "let me check if this works" debugging cycle.

What I'm Most Proud Of

The case study generator — I built it because I needed it. It solves a real problem (context-switching between audiences when marketing your work). It actually generates usable output.

The GitHub integration — Project cards show live data. Not static info I typed in. Real stats that update.

The design system — Editorial, intentional, purposeful. The colors aren't random. The layout has hierarchy. The animations have reasons. Nothing decorative.

The timeline narrative — Six months of progression. Not a flat list of accomplishments. You can see the arc: first project → hackathon selection → hackathon win → new challenge.

It's shipped — Not a concept or demo. Live on Cloud Run, fully functional, judges can actually use the case study generator.