I m Jinny, the solo founder and developer of Crazy Sajangnim.
The name comes from the Korean sale phrase , or The boss has gone crazy. It is commonly used when a store offers discounts so dramatic that the owner must be crazy to sell at those prices.
That idea became the inspiration for Crazy Sajangnim: one place to discover genuinely discounted designer fashion across Korean and U.S. retailers.
We built something because we got tired of watching great candidates disappear into ATS black holes.
After hundreds of applications, interview prep sessions, resume rewrites, and honestly way too many We decided to move forward with other candidates emails
we realized something:
Most people are not losing because they lack skills. They re losing because their CV never gets properly understood.
Been building Yenta for about a year now. The core idea: you text your vibe on WhatsApp, "something low-key, good cocktails, not a tourist trap," and get back three specific recommendations that actually match. No app, no account, just a message. We're live in Dubai with around 500 users and gearing up for a Mexico City launch. The biggest challenge we keep running into is explaining what it is without making it sound like a chatbot or a Google Maps clone, which it isn't. Curious if anyone here has navigated that positioning problem with a product that sits in a category that doesn't quite exist yet.
Global trade is becoming more complex every year. Customs brokers are managing increasing shipment volumes, changing regulations, and thousands of product classifications while still relying on time-consuming manual processes.
Borderline Genius Inc. helps customs brokers streamline compliance workflows with AI-powered tools built specifically for modern trade operations.
How are you classifying AI systems under the EU AI Act today?
One thing I've noticed is that many teams understand the broad concepts of the EU AI Act but aren't always sure how to apply them to a specific AI use case.
So I built a small side project to help.
It's a free interactive EU AI Act Classifier that walks through a series of questions and provides an indicative classification to help start governance discussions.
One thing surprised me after launching Velrix. The most valuable feedback didn t come from people telling me what they liked. It came from people pointing out everything that wasn t good enough. Over the last few days I ve rebuilt a huge part of the audit engine: Modern websites are now analysed the way visitors actually see them. Findings quote the real content on the page. Fixes include concrete rewrite suggestions instead of generic advice. Audit intelligence now flows all the way through to the recommendations. Still building. Thanks to everyone who s taken the time to test it and leave feedback it genuinely shapes where Velrix goes next.
Every AI app I've built ended up with the same screen. Paste your API key, we'll store it. I never liked shipping that, and users don't love it either.
So I built ai-oauth-sdk. It lets someone sign in with the AI plan they already pay for, using the providers' real OAuth flows, the same ones their official CLIs use. The token stays on their device, so you never hold a key and you never pay for their inference.
Eight providers today: ChatGPT, Claude, Gemini, Grok, Copilot, Qwen, OpenRouter and Azure. Zero dependencies in the core package, MIT, and it runs in Node, the browser, React Native and the usual frameworks.
The OAuth was the easy half. What took the time was the things nobody documents. OpenAI's CLI client id is registered for one exact port, localhost:1455, and if it's busy the flow just fails with nothing useful in the error. Cross-Origin-Opener-Policy: same-origin quietly severs window.opener, so a popup callback completes and the parent never hears about it. Google validates redirect_uri strictly enough that a hosted callback is rejected outright, so the loopback URL has to be copied back by hand.
