Skip to content

README.md

Latest commit

 

History

History
91 lines (68 loc) · 4.59 KB

README.md

File metadata and controls

91 lines (68 loc) · 4.59 KB
Raw

Data + Agent Hackathon: hello world

Data Streaming Summit 2026 · a hands-on course in five short labs

You build an agent whose context is a live Kafka stream, kept fresh by streaming SQL, and that asks a human before it acts. Five labs, each adding one idea.

The story

Aegis Financial, a fictional bank, streams every login attempt into Kafka. Somewhere in that stream, an attacker is guessing passwords. Your agent spots them from live data, and flags the account once you say so.

flowchart LR
    K["Kafka topic<br/>security.login_events"] --> S["Streaming SQL<br/>materialized view<br/>login_failures"]
    J["inject<br/>(you, in Lab 3)"] -- "new login burst" --> K
    S -- "SQL tools, over MCP" --> A["Orca agent<br/>hello-agent-&lt;you&gt;"]
    A -- "insert<br/>(only if you approve)" --> F["table<br/>flagged_accounts"]
Loading

Pick your course

The same five labs, on two stacks.

Cloud course Local course
Runs on StreamNative Cloud: your own instance, with a Kafka cluster, a SQL workspace, and an agent workspace Your laptop: Ursa for Kafka, RisingWave, and the Orca Agent Engine (ork local)
You need A StreamNative Cloud login from the hackathon organizers, with a team environment they created or an instance of your own Docker and an Anthropic API key
Time About 40 minutes About 45 minutes, plus image downloads
Start Lab 0: Set up, or from a team card Lab 0: Set up

At the hackathon, take the Cloud course: see Before you arrive. Without a StreamNative Cloud instance, or to see every part run on your own machine, take the Local course.

Pick your path

The agent steps work three ways. Pick one.

The labs

Lab You The idea
0. Set up Get your stack ready and run the doctor Know that every part answers before you build on it
1. Hello, agent Create an agent and chat Agent, environment, session, events
2. Hello, streaming SQL Build a materialized view over the topic Context that keeps itself fresh
3. Agent + live context Give the agent SQL tools, inject new data The answer changes with the data
4. Agent acts, human approves Let the agent write, with your OK Governed actions

Every lab is steps you can check, a short quiz, and a task to try on your own. The labs explains how a lab and its checks work.

Learn with a tutor

A coding agent such as Claude Code can walk you through either course one step at a time, check your work with you, and quiz you. The tutor skill ships in this repository: see Learn with the tutor.

What you build

  • A stream (Kafka) that holds the facts as they happen.
  • A materialized view that keeps a running summary: the agent's always-fresh context.
  • An agent that reads that context itself, through an allow-list of tools.
  • A human approval gate on the one action that changes something.

That is the shape of most data + agent applications. Swap the topic, the view, and the action, and you have your own project: Go further.

What's in this repository

Path What
labs/ The two courses: Cloud and Local
agent/ The agent definition for each lab, per stack, shared by all three paths
sql/ The SQL for Lab 2, per stack, plus a reset script
cli/ The CLI path (ork + jq)
python/ The Python path, the doctor, the seeder, and the data injector
typescript/ The TypeScript path, the doctor, the seeder, and the data injector
local/ The Local course's stack: a Compose file and four helper scripts
lab-ork ork with your endpoint, key, and ids filled in: what the lab checks use
data/, schemas/ The synthetic login events the Local course loads, and their Avro schema
skills/ The tutor skill
docs/ Before you arrive · Learn with the tutor · Go further

Licensed under Apache 2.0.