
I wrote up my tips for teaching Python with GitHub Codespaces, based off what I shared during the #PyconUS Education Summit session:
https://blog.pamelafox.org/2025/06/teaching-python-with-codespaces.html
(It's actually a lot about how to set up dev containers with Dockerfile/docker-compose, which is helpful even if you're not teaching)
#PyConUS talks are going up on YouTube now, wooot.
I co-presented the sponsored session for Microsoft with my colleague Rohit, where we walked through a development flow from VS Code to Azure, using a *ton* of VS Code extensions plus the Azure Developer CLI.
https://www.youtube.com/watch?v=k6Vm2hakkV4
Repo:
https://github.com/rohit-ganguly/rag-postgres-openai-python
My first #PyconUS was super fun. I loved how many mini conferences were wrapped inside of it (like EduSummit and FlaskCon), the friendliness of everyone, and how easy is it was to find people to geek out over the same things I care about.
Kudos to the organizers for their great work, and to everyone who came and brought their friendly curious selves!
Llama ears popped up on another speaker at #PyconUS, what are the odds!?
Want a visual exploration of vectors?
Check out the full size poster here:
https://tinyurl.com/visual-vectors-poster
Repo with visualizations and learning links here: https://github.com/pamelafox/vector-embeddings-demos
Want to geek out on vector embeddings?
Come check out my poster at #PyconUS, from 10am-1pm today!
Spoilers: Here's the digital version for anyone not here:
https://tinyurl.com/visual-vectors-poster
We've put an open space on the calendar at #PyconUS to talk about evaluations for LLM-powered applications!
Come chat about what you want to see in the eval space. What process would help you evaluate your app correctness, safety, and quality?
Sunday, 1pm, Room 319
Thanks to the RAGAS crew for organizing!
Look at the amazing earrings I won at the PyLadies #PyconUS Auction!
1,674 glass beads arranged to represent a Python program.
Should I wear them to my poster session for everyone to check out?
I stopped by @pydantic AI booth to say hi, and it turned into a debugging session with @samuelcolvin to figure out when/why SLMs fail to handle agentic function calling flows. Determined that SLMs like llama3.1:8b can handle a single tool call, but they don't handle a multi-turn conversation with previous tool call responses very well. Reliability drops drastically.