Josh Rufer - AI Developer & Digital Artist

The Agentic AI Developer:
Learning from Pair Programming

Pair programming is a paradigm where two developers share a single computer and work together to solve problems. One developer writes code while the other focuses on large-scale concerns and architecture that may come up in the design. This "two heads are better than one" approach lends itself perfectly to Agentic AI development methodologies.

My over ten years of working in pair programming environments has greatly helped me in transitioning from a solo developer to the orchestrator of an AI coding swarm. By letting AI write boilerplate code and basic functionality, I am free to focus on application architecture and solving more complex problems.

Agenticly Generated Apps

Translate This

Have you ever been traveling and needed to quickly translate something from a language you don't speak, but your internet connection is spotty or roaming charges are just too high? Translate This has you covered. Translate This is a Flutter app for IOS and Android that uses a local on-device LLM to translate between 60 different languages. Translate This doesn't just give you the word you're looking for, it tells you what it means in that language and, in some cases, how to say the same thing in a formal way or a casual way. Because it uses an LLM, it doesn't simply translate your words verbatim. It looks at what they mean. What is your sentence actually asking or saying? It translates that intent. Because Translate This uses an on-device LLM, you can easily translate from one language to another no matter what your connection is like. This architecture also ensures your translations are completely private and secure, never leaving your device.

GGUF Converter & Quantizer

GGUF Converter & Quantizer does just what its name suggests. It can easily take any Safetensor file on Hugging Face and quickly convert it into a .GGUF file and quantize it to any size of your choosing. Simply provide the name of a Hugging Face repository and select the quantization you would like to apply. The app will download the model and begin converting it right away. The result is a smaller model, optimized to use less resources when run. This is perfect for taking a large model and making a small variant that can fit in memory on a PC or even a mobile device. I used this application to quantize models for use with Translate This. I was able to shrink the Gemma 4 model to just under 3 Gig by quantizing it to a 3 bit K_M .GGUF file.

Whisper-Wayland

Whisper-Wayland is an offline dictation application for Linux systems running the Wayland compositor. With it, you are just a keypress away from dictating your text in any text area, in any app on Linux. Using a small on-device LLM, Whisper-Wayland can convert your speech to written text with high accuracy, and security. No need to send your data to the cloud for processing. This on-device architecture ensures your transcriptions are completely private and secure, never leaving your device. Just hold your configurable hotkey and speak into your favorite microphone and the app will do the rest.