Describe a business, get a fully deployable Databricks demo.
Caspers is an AI coding agent skill that builds coherent streaming data, Spark Declarative Pipelines, AI agents, and apps from a natural language description. Say "convenience store" or "craft brewery" and get seed data with PK/FK constraints, a streaming event simulation, and infrastructure-as-code — ready to databricks bundle deploy.
git clone https://github.com/nkarpov/caspers.git
cd caspers
# Open in your AI coding agent (Claude Code, Cursor, etc.)
# Run: /build-business <your business idea>- Describe your business — the skill asks clarifying questions and builds a Blueprint
- Data flows first — seed tables (managed Delta with PK/FK) + streaming events get generated and deployed
- You verify — sample DBSQL queries let you inspect the data before moving on
- Build on top — choose your next layer: SDP pipeline, AI agent, app, or more data
- Stay coherent — the skill validates cross-layer contracts at every step
The repo starts as a skeleton and becomes your business.
| File | Purpose | Maturity |
|---|---|---|
BLUEPRINT.md |
Living business blueprint — tracks all decisions | Core |
databricks.yml |
DABs infrastructure declaration | Core |
data/seed_generator.py |
Managed Delta tables with PK/FK constraints | Full |
data/canonical_generator.py |
State machine event dataset with GPS routing | Full |
data/replay.py |
Streaming replay engine | Full |
pipelines/transforms.py |
SDP bronze → silver → gold | Planned |
agents/agent.py |
MLflow agent definition | Planned |
apps/app/ |
FastAPI + frontend | Planned |
Full = curated patterns, reference code, worked examples. Planned = the skill generates these using the blueprint and LLM general knowledge, but no specialized recipes yet.
┌→ SDP (Spark Declarative Pipeline)
Data (required) ────┼→ Agent (AI reasoning)
├→ App (UI + database)
└→ More data (documents, additional datasets)
Data is the prerequisite. Everything else branches from it in any order.