DataTalksClub

Want to talk about data? Data science, career questions, practical machine learning, and more? Join @DataTalksClub! The link is in the profile description. Learn more about our activities in the thread! 🧵

The new cohort of AI Dev Tools Zoomcamp 2026 starts on August 31. Learn how to use AI coding assistants, agents, MCP, tests, CI/CD, deployment, and automation in practical software development workflows. Register: courses.datatalks.club/register/ai-de… Workshop 1: luma.com/lmkti8zj

This week in LLM Zoomcamp: Module 5, Monitoring. Offline evaluation helps before deployment. Monitoring helps when real users start interacting with the system. Materials: github.com/DataTalksClub/…

Our next live podcast episode is right around the corner! Join us as we chat with @pauliusztin_ about Engineering Your Own AI Assistant. You won't want to miss this one! Register here: luma.com/s5snn8xy

Then we visualize the system with Streamlit and Grafana dashboards and run the stack with Docker Compose. Course: github.com/DataTalksClub/…

We store conversations in PostgreSQL, collect thumbs up/down feedback, and add automatic relevance checks with a built-in judge.

Tracking and personalizing AI agents with Snowplow and the Vercel AI SDK July 21 4:30-6:00 PM GMT+2 YouTube Register: luma.com/vz9iv4f2 5/5

We will configure Snowplow Signals to compute live session attributes from the event stream. Then we will inject those attributes into the agent's system prompt with the Vercel AI SDK. 4/5

On July 21, we are hosting a hands-on on tracking and personalizing AI agents with Snowplow and the Vercel AI SDK Register: luma.com/vz9iv4f2 1/5

Orchestrate LLMs and Agents with Airflow A technical how-to for getting AI into production on Airflow (not another AI thinkpiece). In this 60+ page guide, we've compiled everything we've learned about running AI workloads with Airflow. Download now: astronomer.io/ebooks/orchest…

For agents, we also need to save the final answer and the tool-call trajectory. Otherwise we cannot see which tool calls led to the result. Course: github.com/DataTalksClub/…

Then we evaluate RAG answers. We generate answers for the test questions and use LLM-as-a-judge to measure answer quality.

This week in LLM Zoomcamp: Module 4, Evaluation. Once retrieval and generation work, the next question is: how do we know if they are good? Materials: github.com/DataTalksClub/…

Materials: github.com/DataTalksClub/…

Production checks: monitor tokens, cap output size, store keys as secrets, enable logs, add retries and alerts, document behavior. Good orchestration is also operation.

AI orchestration does not mean every workflow needs an agent. A useful habit from the Kestra module: choose the simplest pattern that fits the problem.