Tue, Oct 28 · 6:00 PM EDT
PLEASE REGISTER ON LU.MA SO WE KNOW HOW MANY PEOPLE TO EXPECT: https://luma.com/ug69dcx6
Come see our awesome lightning speakers!
Speaker: Nick AshworthTitle: Using Local Language models to Increase Accessibility for Offline SensorsAbstract: A live demo of showing how small language models can be used to allow all users of offline mesh networks to easily access and understand both simple and complex sensors. We'll be demoing this on the DC Meshtastic mesh and using it with air and ground traffic sensors.
Speaker: John SobanskiTitle: Building a Machine Learning Algorithm from ScratchAbstract: John shares his experience building a machine learning algorithm from scratch. He outlines the core idea, walks through the math, logic and implementations, and addresses challenges in feature engineering, hyperparameter tuning, and debugging odd behavior. He shows quick runs in Octave, R, Pandas, and Polars, pointing out lessons from each environment. His goal: spark new ways of thinking about the algorithms we use every day and what becomes clear when you work through the details yourself.
Speaker: Jonathan ConradTitle: AI for Analytics: How to Integrate Gen AI into Your Next Data ProjectAbstract: Gen AI is often touted as a miracle solution to data-related problems—but results often end up short of expectations. In this talk, we will discuss how to implement commercial AI models into your data projects to drive superior results at a fraction of the cost. This talk centers around a recent project where we used OpenAI's GPT-4o-Mini model to scrape and classify data from a series of unstandardized documents. I will discuss the general architecture of building a data pipeline that performs these operations, the labor and computation costs, and the best practices we discovered from scrutinizing and refining the process.
Speaker: Abigail HaddadTitle: Hot Claude Summer: My Three-Month Adventure With Claude CodeAbstract: I used to envy developers because they could build websites someone might want to use. But this summer, I was able to use Claude Code to layer data pipelines and web development onto my data science skills—and suddenly found myself debugging at 3 AM after realizing my pipeline was silently deleting data. I'll talk about how more time fixing is currently the cost of using AI-assisted coding, how you can set up a real test/prod framework for free for your data pipelines via GitHub, and about what I still do from scratch.
