My First Pull Request

My first ever pull request on GitHub. And guess what? The hard part wasn’t even the pull request. It was everything before that: making a GitHub account, setting up my terminal, getting yarn, node, git, and Claude Code installed. Figuring out what Claude Code even was. None of those were steps I had ever thought about, but here I was, one day tasked with getting my environment set up.

Once my environment was ready, I got it to push a pull request for me. Not too sure how it worked at first and I couldn’t tell if it had really pushed my changes, so I went and looked at GitHub. There it was. The full description and write-up of everything. It was something I did not think I could pull off, but what I saw was undeniably me and mine.

My first real task for the summer was to run the KORA benchmark. KORA is a third-party open-source benchmark that tests how an AI chatbot responds to kids. More specifically, it’s a three-turn scenario set, over 700 scenarios, where an AI chat model simulates a kid talking to our bot and an AI judge scores what our bot said back.

Like the pull request, my difficulties weren’t with the main task, but with everything around it. Getting the benchmark to run on my laptop and getting the results out of it turned out to be two completely separate problems. The first task was to get the scenarios running, and that was relatively easy. But once it ran, I hit the second problem immediately: I couldn’t read the data.

This was one of my first hard lessons with this experience. Data that nobody can read isn’t worth anything. And work that isn’t committed or published can’t be seen. So I branched off the repo, made a branch for the results, and turned them into tables. Overall score, then the individual categories.

Once I could actually see the numbers, I started to try and improve our score and display it somewhere visually. I worked on how our AI responds in prompting, and what the bot does when a kid says something serious. What happens when a user says I love you, or I feel lonely, or I want to lose weight. This is not a hypothetical for a product built for kids ages six to seventeen, and getting it right matters more than anything else we score on.

And every change meant checking in with Jake, who is the founder of MyDD.ai. So I iterated again and again, and put all of it on GitHub.

Then Jake asked me to display the data. A place where all the information from the KORA benchmark could live and be visualized: current runs, previous runs, and competitors’ runs. It took a few days and lots of iteration to get it right. But it’s done. And now I am able to show Jake our data against other models on the KORA benchmark like Claude Haiku 4.5. I also put in the individual risk categories, as well as a bar chart that puts specific behavior categories directly head to head between MyDD and Haiku.

This was the most important thing I did all summer.

Somewhere in the middle of all this, Jake took me to a conference in New York.

Ah, now I remember, it was actually the start of when I first met Jake in person. He made a joke about us upgrading our relationship to taking trips with each other. It was my first time in New York, and I filmed the whole trip. From taking the Amtrak for the first time, to almost missing the train coming back. It was a whole experience. Not to forget that Jake and I got out of our Uber and ran a full mile through Manhattan to catch our train back! I never thought I’d have to sprint a mile in NYC during an internship, but I pushed past it.

If you are interested in viewing clips from the conference on that day, check out our instagram:
https://www.instagram.com/mydd.ai/

Speaking of social media, I also got to meet the team in person, Jake and Savannah, at The Hallwayz, and we recorded even more content while we were there.

I never realized how much getting a company’s name out there depends on engagement. Being on TikTok and Instagram, being seen, and having people interact with you. That’s exactly how a small company gets known. Once I understood this workflow, I started to think about what else we could do.

So I brainstormed a random idea in a Google Meet with the team, and prototyped from there. The idea was to leave something physical at a coffee shop, and I wanted it to belong there, something that would sit naturally on a brown coffee shop table instead of looking like an ad. I picked complementary colors and designed something that felt like what we’re building at MyDD, then built it with Claude. I acted as the graphic designer, took feedback from the founder, and made it real.

Now we have a flyer that works online and that we can also hand out to somebody in person.

I also worked on MyDD Learn, our AI literacy game. I used OBS and CapCut to make content for it, and I playtested it and gave feedback to the team. Things like better color to keep kids engaged and making sure there were no bugs.

When I was giving feedback, Jake also said something to me. Instead of just providing feedback, I could directly work on the game itself behind the scenes. I never got to that point, but that’s where I’d want to go back and gain more experience.

I came into this summer having no clue how to navigate GitHub and having never designed anything. I’m leaving able to work in GitHub, and with the thinking foundation I need for design. Plus the social media, the flyer, the game, and a mile-long sprint across Manhattan.

But the thing I’ll actually remember is the people. Jake is my founder, my boss, and at this point kind of a friend too. Not really sure what the word is. I just know a year ago I couldn’t have told you what a merge conflict was, and this summer someone handed me real work and trusted me to figure it out.

Until next time,

Signing off, Jack Luong.

Jack Luong is a senior at Penn State and an AI Research and Marketing Analyst intern at MyDD.ai.