It all started with a sticker.
Two years ago, I saw an MLH sticker on a friend's laptop 💻, looked up what it meant, and happened to find Hacktoberfest 2024 running at the same time. That month led to my very first DEV post, My First Dance with Hacktoberfest, where I wrote about submitting my first four small pull requests (PRs).
After a lot of ups and downs, this October looked completely different. Instead of pull requests, Hacktoberfest 2026 asked us to build new projects from scratch using open-source AI and open-weight models 🤖.
Over four weeks, I built five projects (one of which I scrapped hours before the deadline), finished every weekly theme for the completionist++ badge, and discovered that a model I built actually lost. Focusing on research and honest testing is what got Second Look named an Overall Winner in Week 3 🏆.
Quick hello if we have not met 👋
I am Hotragn Pettugani. I have a background in data science and AI, and I love building end-to-end AI products and working with new tools that show their work.
Until the start of 2024, I worked mostly on machine learning, NLP, deep learning, and computer vision foundations, not pull requests or full-stack apps.
If you are staring at your first repository or wondering whether your ideas are good enough to ship, this might help you too. Here is what changed for me between 2024 and 2026, and a few lessons you can borrow for your own projects 🚀.
2024: When a four-line docs fix felt huge 😅
In 2024, I was genuinely sceptical. I had never interacted with open-source tools, libraries, or repos, which made me wonder: is this something I can really do, and is it worth my time?
I expected my PR to be ignored or rejected. Instead, the maintainer left a really kind reply ❤️. I finished October 2024 with four small pull requests. Back then, it was all new to me, and honestly, I was mostly excited about the swag.
2026: Four weeks, four builds 🛠️
This year, instead of chasing swag, I focused on solving real problems. Each week had its own theme. Here is what I ended up shipping:
| Week | Theme | Build | What It Does |
|---|---|---|---|
| 1 | Touch Grass 🌱 | Flare | A social app that lights up when a nearby friend sends a flare, then guides you to them. |
| 2 | Side Hustle 💼 | Cubby | An AI logging assistant for home daycares: caretakers talk, and Gemma writes the daily parent log. |
| 3 | Slow and Steady 🐢 | Second Look 🏆 | (Week 3 Overall Winner) A health plan tool that tells you when the lazy choice is actually safe. |
| 4 | Ghost in the Machine 👻 | Paper Ghosts | A magic-lantern search across 370 million old newspaper articles to uncover the stories behind neighbourhoods and how papers reported them. |
The project I threw away in Week 2 🗑️
The table above looks clean, but Week 2 was a mess behind the scenes.
I originally built Stitch Math, a knitting pattern checker built on Gemma 4. I was really proud of the technical work. Then, right before the deadline, I looked through the community submissions and saw that someone else had landed on a very similar idea 😬.
I wanted my submission to feel distinct, so I scrapped it, started over from scratch, and built Cubby.
That night taught me a big lesson 💡. Everyone gets the same hackathon theme, so it is natural for great minds to think alike. After Week 2, instead of starting from the theme, I started from a real person's daily problem. Ideas overlap all the time, but real problems rarely do.
Week 3: Winning with a loss 📉➡️🏆
Going into Week 3, I expected Second Look's headline to be that my model picks better 2027 health insurance plans than humans do. Instead, my model lost to the laziest rule of all: just take the plan with the lowest premium. So I made that loss the headline and rebuilt the tool around it.
In 2024, I was always focused on making a model perfect and accurate. I probably would have kept tuning until my model came out on top. Over time, working on multiple projects, I realised that failure was the most useful part. If the lazy choice works most of the time, people do not need AI to make the whole decision for them.
So I changed the product to answer that exact question and put the loss right in the title of my post. Being honestly wrong, and building something that shows when a choice is safe, ended up winning the week 🎉.
What I would tell my 2024 self (and anyone starting out) 🧭
If you are trying to build your first project or make your first contribution, you do not need to know everything before you start. Here are four habits that helped me the most:
- Use AI like a team of two or more, depending on the task 🤝. When I used Claude Code this month, I ran two separate sessions at the same time. One acted as the "manager", holding the plan and reviewing code. The other acted as the "builder", writing one small piece at a time. Keeping planning and coding separate kept each session focused, so I could brainstorm, fix, and iterate.
- Start with a person 🎯. If you do not know what to build, ask a friend what part of their job involves annoying guesswork or repetitive typing. Build a tool for that one person.
- Research first, and decide how you will score your work before you test it 📊. Do your research before starting the actual work. At the end, evaluation tells you how well a product or model really performs. That is exactly what revealed the loss behind Second Look.
- Small first steps count 🌱. Build in phases, check every output, and keep testing and refining until you get the result you need.
Congrats to everyone who shipped something this month, and thank you to DEV, @mlhacks, and the community maintainers who make everyone feel welcome 🙌.
Over to you 💬
I am trying to meet more builders here on DEV and learn what everyone else is working on, so I would love to hear from you in the comments (about Hacktoberfest or your personal projects):
- If you are just starting out (or remember when you did), what was your very first open-source fix or side project?
- What is a project you built recently that you are proud of? Feel free to drop a link to your post or repo below 👇. I am setting aside time this week to read through and check out everything shared in the comments!
Top comments (0)