How Jordan built Ryz Score, an AI resume screener that rates applicants against the job description, plus our odd ChatGPT uses for food, money and shopping.
AI resume screening works best when the model scores each resume against the specific job description, not in a vacuum, and posts that score straight into the applicant tracking system recruiters already use. That is how Jordan Metzner built Ryz Score, the internal tool we showed on Episode 4 of Built This Week. Sam Nadler says it has had a big impact on the recruiting team, accelerating their workflow and making them more efficient.
Jordan says the problem starts the moment a job goes live. A post in a popular area, like software engineer or marketing manager, does not draw five or ten applications. It can draw hundreds. It is common for recruiters to spend less than sixty seconds on a resume, and with 100 or 200 applicants that still adds up to hours of reading.
The bigger issue, in Jordan's view, is quality. A human going through a stack of resumes gets tired, gets lazy or misses an opportunity. It becomes a search for something in a large haystack, and people just don't do a great job of finding the best ones. His conclusion was that AI is a great way to score resumes, and specifically to score them based on the job description. In the demo, one personal assistant role had 112 applicants and a Ruby on Rails developer role had 152.
Watch how Ryz Score rates applicants in Episode 4 (starts at 4:23) · Episode 4 notes and transcript
Ryz Score takes a candidate's application and resume and scores them on four things: skills match, experience match, education match and cultural fit. Those combine into one overall score, such as a 90. Recruiters can filter by location or by score, open the resume, and read notes on why a candidate scored the way they did. Further down the list are candidates with zeros, twenties and thirty-fives who most likely won't be a good fit for that job.
The part Jordan likes most is that most recruiters never open the Ryz Score app. The scores are posted back into the applicant tracking system, with a written summary of why the candidate fits. Recruiters work down the list, reach out to the strongest people and send the lower scores a thank-you message. Sam says the team did a lot of testing to check whether a 92 or a 90 actually felt like a 92 or a 90 before putting more trust in the system.
Sam also sees a benefit for candidates. When a recruiter reads resumes by hand, a rock star candidate can sit last on the list until the first group is already in the final stages. Scoring everyone at once fixes that. See our Ryz Score build page for more on the tool, and our AI in recruiting and HR page for other hiring tools from the show.
Jordan built most of it in Cursor. The front end uses Vite, React, Tailwind and shadcn, which he says is what we use all the time. Supabase handles authentication and authorization and stores all of the data.
Supabase edge functions check for new candidates in the pipeline. When one arrives, the edge function sends the data to ChatGPT with a scoring matrix, the job description, the resume and the full application. The score comes back, gets posted into the app, and from there into the applicant tracking system. Jordan expects this to become standard everywhere:
But it seems almost inevitable that every resume will get scored almost immediately for almost any job you apply to in the future.
— Jordan Metzner, Episode 4
He said it surprised him that AI resume scoring had not shown up more in places like LinkedIn or inside applicant tracking systems.
For the tool of the week we covered less common ways we use ChatGPT. Sam has used apps like MyFitnessPal to track calories on and off, but typing in exactly 1.5 cups of rice or five ounces of grilled chicken was cumbersome. Since July he has tracked his calories and macros in a ChatGPT project instead. He sends a voice note or a picture each time he eats, and it does not even need the barcode. It returns calories, protein, carbs and fat, gives a summary at the end of the day, and suggests what to eat later to hit his targets.
It started learning what he usually eats and suggesting it. It also never tells him no, even when he asks about ice cream. Jordan's takeaway was bigger than nutrition:
One ChatGPT conversation is like a mini application. Right? And now it's not a conversation to you anymore. It's actually like a full on end to end application.
— Jordan Metzner, Episode 4
Sam had two more examples. ChatGPT figured out why one of two identical plants on his patio was struggling, and a week later both were thriving. Jordan photographed the menus and chalkboard specials at his cousin's restaurant, and ChatGPT built a digital menu from them.
Jordan thinks it could. He describes the Bloomberg Terminal as an old, clunky interface where users memorize keyboard shortcuts to get things done. He named Perplexity as the first to step up on financial research, and sees Claude for Financial Services, already integrated with providers like S&P Global, as part of a bigger shift.
But I think we're gonna continue to see the degradation of Bloomberg Terminal's value and increase value in using LLMs for financial analysis.
— Jordan Metzner, Episode 4
Jordan uploads option trades with the prices and the Greeks, tells ChatGPT where he thinks the price will go and by when, and asks what return and IRR to expect. He also used it to compare buying Bitcoin against buying IBIT, the Bitcoin ETF, and says he still needs to do more analysis there. For retirement, he gave it his IRA, his portfolio, his goals and his age, and got what he calls a much better holistic analysis of how close he is to his retirement goals.
Sam says Jordan turned him on to using it for a holistic view of his finances. It generally tried to push him toward a more conservative approach, which he says he is not ready for yet. He does not think it removes the need for professional financial services, but it gives him access to far more information and approaches than he had a couple of years ago. Other episodes on money and markets are collected on our AI in finance page.
Sam used ChatGPT to pick a pool vacuum on Prime Day after his first one died in a single season. Jordan sees OpenAI's move as taking over some of the Google pie: if a customer starts their search in ChatGPT and never goes to Google, ChatGPT will want some of that affiliate revenue. He thinks it could have a big impact on Amazon ads, Google Shopping ads and Shopify, positive or negative depending on how well Shopify integrates.
He also thinks it threatens the listicle business, from Wirecutter to BuzzFeed, because an LLM can rank options by what matters to you rather than to everybody. In his words, personalized shopping with direct links for checkout sounds like a huge opportunity and a step forward.
The episode includes the full Ryz Score demo inside the applicant tracking system, Sam's meal logs, and our take on Grok's new companion mode. Watch Episode 4 of Built This Week, where the full transcript is also on the page.
