Riza

Come get a brew at the coffee bar. It's on us. ??

Trying to scrape the web? LLMs can help you extract data, fast. A lot of commercially-useful data is stuck online. For example: CA provides all current and recently-licensed real estate appraisers (link below ??). 13,000 appraisers are shown on 40+ pages with no bulk download option. Here's how we extracted the data: 1?? We fed sample HTML from the site to an LLM 2?? We had the LLM write targeted extraction code 3?? Riza executed that code securely 4?? We got back clean, structured CSV data ?? The key insight: ??Don't ask LLMs to extract the data directly. ??Ask them to write the extraction code. Then run that code securely on Riza. Why? ?? Reliability: LLMs often miss data when extracting directly ?? Speed & cost savings: Make just 1 call to an LLM to generate code, using a small subset of data Try it yourself! Links to full guide, code, and data in comments. ??

Have more data than data scientists? LLMs can help. Data is everywhere, from personal to business to government. But it's time-consuming to analyze by hand. For example: SF publishes all city employee salaries (link below ???). It's now fast to analyze with LLMs. ? Here's a chart Anthropic's Claude generated using Riza that shows salary trends. ? You can repeat this analysis infinitely over other data. Here's how we made the chart: 1. ?? We fed a few lines of the salary data to an LLM (Claude) 2. ? The LLM wrote Python code to analyze and visualize trends 3. ?? Riza executed that code securely 4. ?? We got back a clear chart ? Here, Riza empowered Claude to run Python code using `pandas`, `matplotlib`, and `seaborn` ? Riza executes Python, JavaScript, Ruby, and PHP securely with any libraries you want to use Try it yourself! Links to full guide, code, and data in comments. ??

Have you wanted to evaluate how well models generate code? Here's a step-by-step guide. Follow these 6 steps to run codegen evals safely using HumanEval and Riza. Want us to run more LLM codegen evaluations??Let us know in the comments!

Introducing: the Execute Function API ???? Now it's even easier to execute LLM-generated code with structured inputs and outputs. ?? To get started for free, see the links in the comments. ?? ?? What's new? ???From the start, our original **Execute Code** API has let you safely run any untrusted code. "Untrusted" means code written by LLMs (or humans you can't fully trust, like your users). ?? Our new **Execute Function** API gives you the same safety as before—with better ease of use for calling functions. ?? How it works: ? Pass an LLM-generated function to Riza's Execute Function API ? Send structured inputs (e.g., { "name": "Andrew" }) ? Get structured outputs (e.g. { "greeting": "Hello, Andrew" }) ?? Why use this? → More reliable and debuggable function execution → Empowers you to build coding agents, AI workflows, and data tools → Available via API and SDKs (JavaScript, Python, Go) ?? Full example in the slideshow:

New platform integration guide for Groq and Riza walking you through how to use "tool use" to execute LLM-generated Python. https://lnkd.in/gQ7n6r7D

We're now on YouTube, thanks to Greg Baugues! First video is an exploration of Anthropic's new computer use reference implementation. https://lnkd.in/g6mqaFZS

The API playground is back. Get a feel for how the API works without signing up. https://riza.io/playground

If you're applying for the Mozilla.ai Builders Accelerator and need a locally-running isolated Python or JavaScript code interpreter to run LLM-generated code, let us know in Discord and we'll get you set up. https://lnkd.in/gK8FM7-i

Some really fun stuff coming soon.