Activity
6K followers
-
Marcellus Pelcher posted this“I did only 10 LeetCode problems and got multiple offers from FAANG!” I feel like these types of posts are often designed to grab attention. In one Reddit post like this, the person had also done work on Codeforces, had a friend group of competitive programmers, and had a STEM related degree, all of which likely helped. Study smart by focusing on understanding, not just memorizing problems. But don’t get discouraged when someone says it was easy for them. They may not be telling the whole story. Someone who did 500 problems and failed is not automatically worse at problem solving than someone who did 50 problems and passed. There could have been many other factors at play: prior experience, interview difficulty, interviewer style, referrals, and luck.
-
Marcellus Pelcher shared thisI started learning Rust again. There is definitely a learning curve, but this time I got further than I did in my last attempt. Concepts from The Rust Programming Language book started to click. Using an LLM to discuss the book while I read it was a wonderful learning experience. When it came time to building something with Rust, then I felt my learning was degraded using an LLM. I built one of John Crickett coding challenges in Rust: Pong with Leptos, WebAssembly, and SVG rendering. I think the LLM got in the way of learning. It was great for helping me choose a stack, answering questions as I built, and reducing a lot of manual work. For example, I gave the LLM the image from the Pong challenge page and asked it to create an SVG game board. That saved a huge amount of effort compared to UI work I have done in previous games. But the coding felt... too easy. At times, I felt more like a coach on the sidelines than the person on the field. After getting the game running, I walked away feeling like I gained only a small fraction of the skill I would have gained if I had built it line by line myself. I write most of my code with an LLM at work, but this was different because the goal was not just productivity. The goal was to learn the language. In that context, I think the LLM took something away from the experience. I did not simply prompt, “Build me a Pong game.” Most of my prompts were smaller and more specific: architecture questions, Rust questions, Leptos questions, rendering questions, and feature-by-feature implementation questions (sample prompts in the comments). But even then, I do not feel like I learned as much as I wanted to. For people using AI while learning a new language: how do you decide when it is helping you learn versus helping you avoid the helpful hard part? Live game: https://lnkd.in/dVdKC2br Source: https://lnkd.in/dAPwcFJQ Controls Player 1: Q / A Player 2: O / L Press any key to start John Crickett Pong Challenge: https://lnkd.in/dDMParxY
-
Marcellus Pelcher shared thisGiven the recent layoffs at companies like Atlassian, Coinbase, and Meta, I wanted to share my own layoff story from earlier in my career in case it resonates with anyone affected. The night before I was laid off, I was on a late-night call with a colleague in Israel, so I slept in a bit the next morning. When I woke up and logged into my computer, I saw the dreaded no access avatar. There had been a layoff meeting, but I missed it. Around 20% of the company had been laid off. I sent my manager a text: “No hard feelings. I wish you and the team success.” Three things I want to share from this experience: First, at the time, it felt like one of the worst things that had ever happened to me. I remember reading posts from people saying their layoff ended up being a net positive in the long run. Honestly, that did not feel true in the moment. But looking back now, I understand it more. A layoff can be painful and stressful, but it does not have to be the end of your career. Second, do not let friends or family project their fears onto you. Your layoff is your layoff. You know your finances, your skills, your timeline, and your options better than anyone else. Listen to advice, but do not let other people’s fear make your decisions for you. Third, reflect on your time there, but accept that you may never truly know why you were chosen. Learn what you can, but try not to obsess over every decision, mistake, or moment you wish you could redo. Reflection is useful. Rumination usually is not.
-
Marcellus Pelcher posted thisNurse: “What are you going to school for?” Me: “Computer Science.” Nurse: “Oh, that’s a bad field to be in if you want a job. My boyfriend was in that field.” This happened while I was getting ready for ear surgery circa 2004. Recently, I heard that computer science classrooms are emptying out. I remember hearing similar things after the dot com bubble burst. A computer science degree is not as fashionable as it once was. In my opinion, if you are only in it for the money, you should probably choose another path. I wanted to be a computer scientist both before and after the dot com boom. Just as I felt then, and still feel now, I think there will be opportunities for people who genuinely care about the field. Software development may change a lot from what we know today, but solving problems with computers is not going away. To me, that is the core of computer science. We do not say astronomy is dead just because telescopes have become more automated.
-
Marcellus Pelcher shared thisUsing LLM to go through diagnostic information is a good use case. I found them to be useful reading SQL query plans.Marcellus Pelcher shared thisClaude Code can help you find optimization opportunities in your Java code. Follow these 3 easy steps: Step 1: Enable Java Flight Recording for your load test or benchmark Step 2: Point Claude to the generated recording Step 3: Prompt Claude to look for events related to your optimization problem The jfr print sub-command comes in handy here since text output is something Claude can process and understand. If you're curious about the capabilities, execute `jfr help print`. It is important to keep a record of your results across your attempts so you can identify when you've actually made improvements or suffered from a regression. Happy hacking. #softwareengineering
-
Marcellus Pelcher posted thisLLMs are not human. It is important to recognize this distinction to maximize their potential. Understanding their capabilities and limitations can lead to more effective utilization.
-
Marcellus Pelcher reposted thisMarcellus Pelcher reposted thisMy advice to young people interested in computer science but worried about AI: https://lnkd.in/gmrYr-tG
-
Marcellus Pelcher shared thisTo effectively work with LLMs require mental adjustments from how you may think will they work and what you think they will be bad at. Also, keep trying to push at the boundaries because new models are better than the previous ones. John is spot on here:Marcellus Pelcher shared this"AI shouldn't review code it wrote." Wrong. The concern makes sense at first: if the AI wrote the code, it'll just approve its own logic. It can't catch its own mistakes. That criticism is completely valid when developers review their own code. Same brain, same blind spots, same assumptions. Self-review has real limitations. But AI is different. We can wipe its memory. Step 1: Generate the implementation. Step 2: Clear the context (wipe its memory). Step 3: Review the code against the requirements and specification. Now the reviewer doesn't know the original intent. It doesn't remember the shortcuts or the "good enough" decisions. It only sees the code and the spec. Fresh eyes, every time. No ego. No attachment to the solution. That's not the same AI reviewing its own work. That's an independent reviewer who happens to share the same capabilities. If anything, this is more objective than human self-review. Unlike humans, AI memory is optional, not permanent.
-
Marcellus Pelcher posted thisA huge chunk good communication is reducing surprises.
Marcellus Pelcher replied to a comment
5d
Adam Wolff My worry is that if I rely too much on the LLM while learning, I may not build the foundation needed to debug when the LLM gets stuck.
I fear the “You are absolutely right, the bug still exists” loop, where it proposes another fix that also does not work.
If I do not understand the language well enough myself, I may not be able to get unstuck.
Marcellus Pelcher replied to a comment
5d
Stephen Hawkins thank you for this insight. Digging into an existing project probably will be better for me.
Marcellus Pelcher replied to a comment
6d
John Crickett thanks! I appreciate it. Before I share it there, I think I will polish it a little more.
-
Marcellus Pelcher reacted on thisMarcellus Pelcher reacted on thisExcited to share that recently I have successfully cleared the interview process for the Google Software Engineer, Early Career role and have reached the Team Matching phase! If you’re a Hiring Manager, Software Engineer or on a team hiring for L3 Software Engineers at Google, feel free to reach out. Ever since I chose Computer Science after 12th grade, working at Google has been a dream. Interviewing at Google’s Mountain View office was one of the most challenging experiences of my career so far, and I’m grateful to have successfully made it through. Today, I feel like I am one step away from turning that dream into reality. Brief intro about me - I recently graduated with an M.S. in Computer Science from Arizona State University. I also recently interned at Arrcus, Inc. in the Bay Area and previously at MSCI Inc. Along the way, I also had the opportunity to interview with Bloomberg, Meta, Cloudflare, Rubrik and Palo Alto Networks, which helped me grow immensely through this journey. Huge thank you to Daniele Souza for the support and guidance throughout the process. Please feel free to leave a comment or send me a DM. #Google #SoftwareEngineer #TeamMatching #L3 #EarlyCareer
-
Marcellus Pelcher liked thisMarcellus Pelcher liked thisFor over a year, I have been exploring computer vision from first principles, delving into the fundamental question: How does vision itself work? My journey began with the structure of the human eye and led me through several key topics, including: - How the retina processes light - How the brain builds visual representations - The significance of convolution in vision - How neural networks learn image features - The impact of transformers on computer vision - How generative models like VAEs, DALL·E, and diffusion models create images I have compiled this learning experience into a free playlist of 30 videos, complete with slides, on my YouTube channel, Code Emporium. The aim is to assist curious learners in understanding modern computer vision by starting from biology and progressing to today’s AI systems. Here is the playlist: https://lnkd.in/g6dgEh_d I will continue to update it as I learn and create more content. If you find it useful, please consider liking the videos to help them reach a wider audience. Happy Learning! ↓ Check out our Youtube channel to understand AI with 155,000 other learners:**https://lnkd.in/gAD7KgKY**
-
Marcellus Pelcher liked thisMarcellus Pelcher liked thisBearly Fit is 1 years old today! Bearly Fit started to help me with my own health. An app that would let me monitor my exercise, nutrition and body in one place whilst still owning 100% of my data. Last year after months of building I finally released and I was absolutely blown away by the response! Hundreds of people downloaded, shared feedback, joined the community and even became paid subscribers. And I've been riding that excitement ever since. In the spirit of building in public - I wanted to share my outcomes. From a developer perspective: - 70~ feature requests - 30~ bug reports - 8 Releases (1 pending) - 2k~ hours - 136k lines of code - 1.3k commits - 1,534 files changed, +100,403 / -24,275 That's just for the app. Most of the code is manually written - AI has mostly only been used for refactoring work and tests. And from the business side - for the first year, I just focused on building. I thought by not investing in marketing or advertising, it would give me and others a good baseline for what to expect. So here's what that looks like: - 2k Downloads - 400 Active users - $900 in revenue - 4.8* and 54 reviews across both stores - 182 on Discord - 2200 followers across platforms This is with the caveat that I have 25k followers and people (for reasons unknown to me) actively want to see me succeed. So even though this is a good baseline, it's still maybe a little inflated. My goal this is to get the remaining features in place with better nutrition logging, wearable integrations, developer features, social features and a web based deep analysis tool. But I'm also going to start to slowly ramp up both marketing and advertising. and I'm excited to see how that impacts it's growth. I hope you've enjoyed reading this and I wanted to say thank you again to everyone that's been part of this journey. I'm excited to see where we will be for our 2nd Birthday!
-
Marcellus Pelcher liked thisMarcellus Pelcher liked thisAs an autistic individual, I need time to think before responding to a question. Not because I'm mentally incapable — but because I'm actively calculating the appropriate response based on the audience I'm talking to, to make sure it isn't misunderstood. And do you know how exhausting that actually is? To analyse your current audience. To gauge what level of abstraction they need so they don't think you're insane. It is exhausting. So being patient — and not assuming we are slow or incapable — is the least you can do, rather than judging it.
Experience & Education
-
Google
View Marcellus’s full experience
See their title, tenure and more.
or
Licenses & Certifications
-
-
-
-
-
-
-
LPIC-1
Linux Professional institute
Issued ExpiresCredential ID LPI000221612 xm5mwdrf3t -
-
Creating a Responsive HTML Email - Lynda.com
-
-
Volunteer Experience
-
Software Engineer
Dallas Code Camp
- Present 14 years 9 months
Science and Technology
http://dallasgivecamp.org/
Publications
Courses
-
Code Eval - https://www.codeeval.com/public/f07238ffc86e5bb3482de02257c42de8b91b2640/
-
-
Code School Courses - https://www.codeschool.com/users/98845
http://tinyurl.com/ljxt3d
-
Pyschool - http://www.pyschools.com/quiz/view_ranking
SysCoder
-
Tree House Courses - http://teamtreehouse.com/marcelluspelcher
-
Projects
Recommendations received
1 person has recommended Marcellus
Join now to viewOther similar profiles
Explore more posts
Explore top content on LinkedIn
Find curated posts and insights for relevant topics all in one place.
View top content