Databricks Oct Meetup
Details
Speakers
Thoughtworks: Eric Nagler and Arun Srinivasan
Databricks: Kevin Hartman
Location
200 E Randolph St. Chicago Floor 25
Hosts
Thoughtworks
Agenda:
5:00-5:30p.m Happy Hour
5:30 (45–60 min) — AI-Ready Semantic Layers: from ontology to Genie
- The problem: metric views give agents dimensions & formulas, not part-of / is-a / rolls-up-to — and Genie Ontology only infers what your assets already encode (the "right answer, wrong rollup" failure — not a model problem).
- The moat: the platform moved the floor; the domain-specific semantic layer (metric views, glossaries, ontology) is still the differentiator.
- The build: rapid discovery with the Ontology Accelerator → where to model non-metric concepts (tables vs. UC semantics vs. ontology vs. agent) → two live demos (infer vs. supply; end-to-end Genie Agent).
6:30 (~60 min) — Consort (open source)
- What it is: keeps AI-written code clean and correct — a deterministic state machine drives /plan → /design → /build → /deploy so agents can't skip steps, weaken tests, or drift from spec.
- Why Lakebase: every git branch auto-pairs with a copy-on-write Postgres branch in ~1s — tests run against real data, not mocks, and schema evolves in lockstep with code.
- How it works: 8 role agents (Spec Author, Architect, DBA, Test Strategist, UX, Navigator, Driver, Product Owner) communicate only through immutable artifacts; RED → GREEN → refactor; human-approval gates that fail closed.
- The differentiator: engineering discipline enforced by contract, not suggestions — "green" means a real test pass on live infra. Live /consort:start scaffolds repo + paired DB + agents (StockFlow sample).