dpguthrie/mastering-ai-observability-workshop — explained in plain English
Analysis updated 2026-05-18
Run a local chat interface backed by a live AI support agent to explore its behavior.
Seed and run an evaluation dataset in Braintrust to measure an agent's quality.
Import sanitized example traces and turn them into draft evaluation test cases.
Follow a live workshop exercise on building an AI observability and improvement workflow.
| dpguthrie/mastering-ai-observability-workshop | 920linjerry-stack/capital-studio | adya84/ha-world-cup-2026 | |
|---|---|---|---|
| Stars | 16 | 16 | 16 |
| Language | Python | Python | Python |
| Setup difficulty | moderate | easy | easy |
| Complexity | 3/5 | 3/5 | 2/5 |
| Audience | developer | researcher | general |
Figures from each repo's GitHub metadata at analysis time.
Requires a Braintrust API key and a chat model that supports tool or function calling to use live features.
This repository is the hands on workshop app used for a session called Mastering AI Observability, presented at the AI Engineer World's Fair using a company called Braintrust. It demonstrates what the README calls a trace-backed agent improvement flywheel, meaning a workflow where you watch how an AI support agent behaves in practice, turn interesting real examples into test cases, and use those test cases to keep evaluating and improving the agent over time. After installing the project's Python dependencies and copying an example environment file, a local chat interface runs at a local web address, backed by a small seeded database. To actually talk to a live AI model through this chat interface, you need your own Braintrust API key and to specify which chat model to use, and that model must support tool or function calling since the support agent relies on that capability. The repository is organized around a set of shortcut commands rather than a single script. These commands can seed an evaluation dataset in Braintrust from a local file of example cases, run that evaluation, push custom scoring definitions to Braintrust, import a shared bundle of sanitized example traces, generate draft evaluation cases from those traces, and write ongoing production style traces back to Braintrust for later analysis. A readiness command checks that the environment, local database, and live model access are all working before starting the workshop exercises. The README itself is intentionally short, since detailed instructions live in a separate workshop guide file and a second document describing a custom classification prompt used in the exercises. As presented here, this is workshop supporting material rather than a general purpose, standalone product, and it does not state an open source license.
A hands on workshop app that demonstrates using Braintrust to trace, evaluate, and iteratively improve an AI support agent's behavior.
Mainly Python. The stack also includes Python, Braintrust, uv.
Setup difficulty is rated moderate, with roughly 30min to a first successful run.
Mainly developer.
This repo across BitVibe Labs
Don't trust strangers blindly. Verify against the repo.