IDEATION
Going into the hackathon, we knew we wanted to focus on the agentic applications category. Part of the goal was to get hands-on experience actually building in this space, and to better understand the growing landscape (and competition) around creating agentic applications today.
During our brainstorming, my friends and I kept circling back to the same pain point: every time we tried to start a new project, we wasted hours hunting down dependencies and downloading packages before we could even get coding. That realization gave us our direction.
TECH STACK & IMPLEMENTATION
We built a command-line tool that asks in plain English what you’d like to build, then uses Ollama’s GPT-OSS model to turn that description into a list of real, installable packages. The tool then handles setup automatically by installing dependencies through Homebrew.
On the backend, we used Python’s Click for the CLI, subprocess for executing installs, and Poetry for packaging the Python side.
We had a lot of difficulty at first, as our integration with the OSS model was painfully slow, sometimes taking close to an hour just to generate a response. We realized the application was too large for our laptops to process efficiently, and debated whether it was worth locally hosting the application or just reworking the platform to use an api, but in the end decided it was too cumbersome to have people go through that intermediate step. We went back in, tested variations, and kept refining our prompt engineering until we were able to cut the response time down to something actually usable. That process ended up being one of the most valuable lessons of the build.
My focus in the tech stack was on the Ruby packaging layer, and I developed the RubyGem wrapper that bundled everything neatly, making the tool distributable and easy to install across environments. That way, what started as a hackathon experiment became a clean, shareable utility that we now rely on to skip the setup grind and get straight to building.
