GPT-OSS Showcase

first page
s3econd page

Inspiration

Honestly, the open-source generative AI world? It’s like a wild sci-fi movie out there. Every day someone’s dropping something nuts with Llama, Phi, Gemma—pick your power model. You can practically feel the brainwaves buzzing through forums and Discords. That energy? It got under our skin. We just had to build a place to show it off, pump it up, give these mad projects their moment in the spotlight.

What It Does

Say hello to GPT-OSS Showcase. Think of it like an arena for coders, where your hacks don’t just get tossed onto a pile—they get spotted, gawked at, roasted, or even cheered by fellow nerds and AI. Platform’s built to spark new tricks, foster some learning (assuming folks are into that), and mostly, let everyone bask in the pure talent bursting out of this community.

How We Built It

We didn’t skimp on the stack—had to keep things beefy:

Framework: Next.js 15 (using that fresh App Router) Language: TypeScript (seriously, fight me—once you go typed, you can’t go back) Styling: It’s all Tailwind CSS, plus ShadCN UI, because let’s face it, no one wants “default Bootstrap blue” in 2024. AI Integration: Genkit handles all our AI wizardry, doing the heavy lifting with Gemini models from Google. Anything from review writing to tagline spitballing, it’s in there. 3D/AR Viewer: For anything hardware or robotics, we’re rocking the @google/model-viewer web component—right in your browser, spin those robots around like Tony Stark.

App’s organized so it doesn’t melt your brain under the hood—UI’s got its own area, AI flows have a comfy carved-out spot in src/ai/flows. Feels like a garage, but for code.

Challenges (Aka, Stuff That Nearly Broke Us)

Biggest beast? Prompt engineering. Turns out, teaching an AI to write reviews that don’t sound like colorless legal letters, and to sum up projects in clever taglines... that’s not a “one prompt and done” job. We redrafted so many times, you’d think we were hunting for the Holy Grail of instructions.

Oh, and keeping things zippy. AI spit out reviews in a couple seconds, but—whoops—the app would just... freeze. Had to work some loading state magic and flash fun little “We’re working on it” messages so folks didn’t think the site died. Don’t get me started on making client-side 3D stuff play nice with server-side rendering—wrangling those hydration errors... chef’s kiss, nightmare fuel.

Achievements Worth Bragging About

Mixing Up the AI Party: We actually pulled off three AI features that aren’t just “here’s some text.” The auto-assessment gives real feedback, taglines actually have a bit of flair, and all of it fits together. It ain’t just fancy autocomplete. 3D That Pops: Thanks to the interactive model viewer, you can now ogle a robot build in the browser. No more low-res screenshots with coffee mugs for scale. It’s like AR Pokémon cards, but for code. Dead-Modern Tech Stack: Next.js 15, React Server Components, TypeScript, ShadCN UI… If a recruiter saw this they’d start hyperventilating from joy. UI That Doesn’t Suck: Behind the curtain, it’s a circus, but what you see is clean, smooth, and snappy. Easy on the brain and the eyeballs.

What We Actually Learned

This whole thing turned into a crash course in reality. Genkit made the AI pipeline... less painful than I feared. Building with Zod to keep data tight. Suddenly, things like:

Auto-grading a project for quality,
Generating artsy taglines,
Shoving 3D models right in the page,

...all felt doable instead of “we’ll get to it next year.” And yeah, Next.js 15 with React Server Components? Fast as hell. UI couldn’t have come together this slick without Tailwind and ShadCN keeping everything on brand. What’s Next For GPT-OSS Showcase

We’re just getting warmed up—seriously. Here’s what we’re hyped about rolling out next:

AI Semantic Search: Forget praying the right keyword is somewhere in a project’s name. Soon you’ll be able to just type “projects that help on farms” and—boom—it’ll find some AgriScan magic for you, even if “farm” ain’t in the description. Automated Content Tagging: So people can actually find stuff (and I can stop answering “Where’s that robot thing?” in Slack every other day).