How Harnesses Improve Model Outputs

Today Quincy Larson interviews Landon Gray. He's a software engineer who worked at agencies for years. Then he taught himself AI assisted software development. And now he's helping other devs do the same.

Landon's famous for proving that RAG pipelines can be written in Ruby and popularizing Ruby as a language for building machine learning projects.

He works as an AI Engineer at a enterprise software company and runs a popular newsletter.

We talk about: - How Large Language Models are just the raw fuel, and harnesses are the real engine to get things done - Why building your professional network is so helpful for finding clients and landing job interviews - Why Landon helped port Python machine learning libraries to Ruby, and why he thinks that – now that AI is just an API call away – the Ruby ecosystem is better-positioned than ever.

Support for this podcast comes from the 10,113 kind folks who donate to our charity each month. Join them and support our mission at https://donate.freecodecamp.org

Get a freeCodeCamp tshirt for $20 with free shipping anywhere in the US: https://shop.freecodecamp.org

Links from our discussion: - Landon's Substack newsletter: https://landongray.substack.com

Community news section:

1. freeCodeCamp just published a new YouTube course that will teach you beginner Front-end Development skills like HTML, CSS, and JavaScript. You can code along at home and build a variety of projects: your own interactive quiz game, a currency converter app, and even a Trello-style kanban board. Along the way you'll learn how to use APIs and local storage to extend the functionality of these bite-sized apps. (12 hour YouTube course): https://www.freecodecamp.org/news/build-19-web-dev-projects-using-html-css-javascript/

2. Learn how to properly test your software and ensure it doesn't break when you add new features. Prolific freeCodeCamp instructor Beau Carnes teaches this course. He'll introduce you to the Testing Pyramid and show you how to balance fast unit tests against complex end-to-end user journeys. You'll also learn how to automate some of this testing using an open source library called Playwright and an LLM testing tool. (1 hour YouTube course): https://www.freecodecamp.org/news/software-testing-with-playwright/

3. More and more apps are relying on probabilistic LLM output alongside deterministic API calls. This makes life harder for devs who now need to ensure that hallucinations don't escape to end users. freeCodeCamp just published this advanced observability tutorial that will teach you emerging best practices and architectural patterns for dealing with this. (40 minute read): https://www.freecodecamp.org/news/build-end-to-end-llm-observability-in-fastapi-with-opentelemetry/

4. Learn how to containerize your MLOps pipelines. This tutorial is the result of hard-won deployment wisdom. The author spent three weeks debugging a Python library error due to dependency conflicts. His eventual answer: containerize entire project with Docker. This tutorial will show you how to structure your containers with multi-stage builds. You'll also learn how to set up experiment tracking with MLflow, versioning with DVC, GPU passthrough, and other advanced techniques. (40 minute read): https://www.freecodecamp.org/news/containerize-mlops-pipeline-from-training-to-serving/

6. Today's song of the week is 2006's Everybody by UK producers Basement Jaxx. If you're familiar with their work, you know you're in for a psychedelic yet silly romp. Between the spoons, bongos, and swooning chorus the song feels like it's held together with duct tape but it works. https://www.youtube.com/watch?v=OrMot81VE8g