Josh Charpentier

Barry the Bench Maker is my most recent project. Originally, I created a basic version of this game in high school with JavaScript, HTML, and CSS. I wanted to garner C++ experience so I revisited it after all this time, while giving it a fresh coat of paint. Barry the Bench Maker is a full-featured terminal RPG. Learning a new language at this point is largely syntactical. I got familiar with C++ for the first time, and really enjoyed taking my time to work on a game. The creativity fostered by this game, even if only existing in a terminal, made it a joy to create. Click play to give it a try!

Thought

I recently finished reading The Pragmatic Programmer by Andy Hunt and Dave Thomas and despite the vast amount of good advice, one phrase stuck out to me: "no broken windows". My understanding is that there was a study done which showed that a building with one unrepaired window soon had multiple unrepaired windows, and other signs of neglect. The idea being that once people saw that no one cared enough to fix the first window, there was less of a desire to maintain the rest of the building, because, again, no one cared about the window. Hunt and Thomas extend this idea this idea to software engineer, suggesting that as software engineers, we should not let neglect slip into our codebases. I've often found myself seeing something and thinking "someone else will get to this" or "this is an issue, but not large enough to warrant addressing". Lately, since I've read this book, I've stopped thinking like that. I remind myself, "no broken windows". Fix the issue now, or at least address the issue in an upcoming PR. It's been my experience that unclean codebases can tend to run rampant when not enough care is put into maintaining them. That care needs to be put in constantly, and truthfully, it requires active decision making and thinking that not everyone wants to engage in. I think in the long run though, the whole team will be glad if they can look at a well maintained and easy to understand codebase. Fix any broken windows.

Like BreakIt and MadCourses, CoverMe came from a personal need. When my school advisor suggested applying to roughly 100 jobs just to hear back from a handful, I knew I needed a way to stand out while staying efficient. Cover letters were a potential differentiator but crafting unique ones for each posting was a long process. CoverMe feeds your resume, a scraped job description, and optional writing samples into an LLM to generate a first draft directly in an editable interface. No prompting, no switching tools. Users can choose between running models locally via Ollama or through OpenRouter in the cloud.

Thought

In general, there seems to be some passive aggressive/ironic conversation around using Obsidian, of which I'd heard little about. After giving various note taking apps a try I decided to try Obsidian as well, and it is really transforming how I learn. I find that I used to be much more passive in my intake of information, whether that be through watching videos or reading books. The fear was that I'd write notes but never again look back at them. Luckily for me, someone created an Obsidian plugin to turn notes into built-in Anki flashcards with ease. This has been a life changer. I have hundreds of flashcards now and each day I have a set of flashcards to look through and practice for spaced repetition. I am feeling the benefits of this more than any past attempts to really get into algorithmic thinking. I should also note that Obsidian's easy linking between notes is an incredibly useful feature. Lastly, I think much of the meme-focused conversation around Obsidian is that the vibe-coding tech bro will link it to their local AI system and have it make all these useless notes. Honestly, I can't attest to AI integration as I haven't attempted it. The same can be said of many other plugins that are made for Obsidian. I really feel like people should give it a try as a note taking app, and if they feel limited, a plugin probably has your back. Add plugins iteratively as to not be overwhelmed.

MadCourses is a tool I built in my final months of college. I was thinking back on what would have made my course selection experience easier. I knew I wanted to study computer science since elementary school, but which courses would actually build the skills employers wanted? I would have loved a tool that matched job skills to UW-Madison courses directly, so I built one. Users enter skills and get ranked course recommendations. Under the hood is a RAG system: course descriptions are embedded and stored in Vercel, and Hugging Face encodes the skill input at query time to find the closest matches via cosine similarity.

Thought

Ever since the term "vibe coding" or "AI-assisted coding" became a real thing, I've noticed two polarizing camps emerge. The first claims that if you vibe code, you're less of an engineer because you don't actually know what you're building. The other believes that if you aren't vibe coding, you are less of an engineer due to your inability to keep up on emerging tech. I've seen very little in the way of a middle ground. In reality, I've found there is a vast middle ground where most of us should be operating. You can use AI to your advantage while still maintaining the knowledge needed to understand what's happening at the code level. Those who I've witnessed seem to make the most of the middle ground are those that not only can understand the generated code, but also know the codebase so well that they can easily pick apart lies from a given AI model. For instance, some model may claim to have the best approach to a solution, and as someone with minimal codebase understanding, you may believe it, whether it's correct or not. This is where the problem seems to stem from: the potential to end up with some faulty product and not know why. I believe the more unfamiliar you are with a given codebase the more you should look to avoid putting your code in the hands of a model before you understand it entirely.

I've always believed in building with purpose; the strongest motivation I have to create something is when I can solve a problem for both myself and others. BreakIt was born when my dad pointed out an anxious habit of mine: picking at my nails. I thought, there has to be a way to track this and stay accountable. It was my first React Native app, which gave me hands-on insight into cross-platform, full-stack development. I learned much about state management and app deployment thanks to this project.

Thought

Sometimes I have doubts about how I am attempting to grow my knowledge and position myself better for the future. There is, after all, so much to learn when it comes to computer science. Each subsect of the industry has infinitely more to learn, and it can sometimes be tough to feel that I am doing enough. The added layer of this being *drum roll*: AI. Will some of what I'm learning be fruitless if we delegate certain tasks to AI in the next few years? Why learn C++ and how to create video game art, for instance, if AI will be doing those things before too long? Everyone, even you reading this, likely has an immediate answer when I pose those questions, but the truth is none of us know how the future is going to play out. While it would be foolish to ignore all the opinions and predictions, they cannot be taken as law either. At the end of the day, I need to focus on the tools I have at my disposal and how I can utilize those tools while also staying focused on my goal. Yes, AI exists, and it won't stop me from learning things that may or may not be useful in 5-10 years. But is usefulness the only reason to learn something? I would argue that the mere act of learning is at least equally important. Pushing oneself to grow and adapt to new and changing landscape will only ever be a good thing. In the meantime, I've found I will utilize AI to help me learn. Here are some of the ways I'm already doing this: - Detailing project folder structure + what each file will do according to best practices - Spelling out project requirements based on an idea + back and forth feedback - Creating individual file templates (commented up boilerplate including methods for me to implement manually; comments may include topics to research too) - Deliverable art requirements based on the scope of the project and more. This is all to say, I can't predict the future, I need to stay focused, and remain confident that investing in myself with the tools available will get me to where I want to go.

Thought

Recently, Anthropic announced that they would be giving their latest model, Mythos, to a select number of companies rather than the general public due to its prowess in finding dangerous security vulnerabilities. Something that I found interesting surrounding the discussion of Mythos and its implications, was the idea that more modern languages are sometimes being encouraged in opposition to languages that allow memory manipulation and other low level capabilities like C and C++. I wonder what the implications of this would be in certain industries. As far as I am aware, the reason so many video game developers rely on C++ is because it allows for greater efficiency thanks to, in part, the ability to manage memory manually. It seems like the strength of a language like C++ is also evidently its weakness. I wonder if AI models will find that these lower level languages are much more easily exploited than modern languages. Only time will tell. Furthermore, it's tough to tell if many companies really are trying to steer away from these lower level languages or if this was an overstated occurrence. My inclination is that, in reality, the number of companies making the decision to avoid certain languages due to security vulnerabilities is very few, both because of the different needs for these lower level languages as well as the feasibility of switching at this point. On a team that primarily programs in C++, converting a codebase to a more modern language not only would take months if not years, but also would rely on the ability of many to potentially have to learn a new language as they convert code over. I've seen this on a smaller scale in the workplace and it seems like it's doable but may limit the ability of the team to make informed decisions. On the flip side, it could be argued that the investment of adapting a new language is less and less as AI models do more and more of the actual coding. Food for thought.

Thought

I am nearing the completion of this new website redesign! A few things to highlight: now things are in more of a feed, and as you may notice, there are thoughts like these that pop up now and again! I think a lot about computer science and the tech space and realized it would be fun to have a place to put some of these thoughts. What better place than my own website? The feed isn't strictly chronological - more aimed at balancing thoughts with projects. Use the filter to see one or the other. The "All" page also may sometimes have a featured section like it does now! Check out my latest project and first game: Barry the Bench Maker! (you can play it right on this website). And lastly, as always, we have my Github, LinkedIn, and Resume links in the bottom right. Thanks for stopping by and feel free to reach out on LinkedIn!