Lakehouse Prepchief architect

#Databricks Chief Architect — Interview Preparation

A complete, self-contained preparation program for a Chief / Principal Architect role centred on leading architecture across enterprise Databricks engagements.

Built for a candidate who is solid but rusty in places, with an interview inside 1–2 weeks.

#The role this targets

From the role description supplied:

  • Lead architecture across enterprise Databricks engagements
  • Design secure, scalable and high-performing Lakehouse platforms
  • Deep expertise in Apache Spark, Delta Lake and Lakehouse architecture
  • Remain hands-on with Spark performance, debugging and optimisation
  • Deliver Databricks solutions at enterprise scale

Two interview archetypes are covered, because the title appears in both worlds:

Archetype Where it lives What the loop optimises for
A. Field / delivery architect Databricks Field Engineering, or an SI/consultancy Databricks practice Customer discovery, the panel presentation, objection handling, depth under pushback
B. Enterprise chief architect A customer building Databricks as the strategic platform Target-state architecture, TCO and business case, governance, stakeholder influence

Most senior loops blend both. The material is written so you can run either, or both.

#How to use this repo

If you have ten days, follow 00-plan/03-ten-day-sprint.md literally. It sequences every file here into daily blocks and tells you when to stop reading and start drilling.

If you have three days, do this and nothing else:

  1. 99-reference/cheatsheet.md — one pass, out loud
  2. 01-platform/06-performance-and-cost.md — the Spark diagnostic playbook
  3. 02-drills/05-the-panel-presentation.md — build the deck, rehearse twice
  4. 02-drills/01-architecture-design-scenarios.md — whiteboard three scenarios on paper
  5. 03-practice/01-mock-loop-full.md — run the mock, score yourself

If you have one evening: cheatsheet + presentation outline + the top-20 flashcards.

#Read it online

#→ lakehouse-prep.pages.dev

Live on Cloudflare Pages. Everything in one place:
The handbook All 26 chapters, searchable (press /), with sidebar and on-page contents
The drill deck Sprint tracker, 87 flashcards, timed drills, cheatsheet
Single-file copy The whole handbook in one 630 KB page — save it, works offline

Also published as private pages: handbook · drill deck.

The Pages site is public to anyone with the URL. To lock it to your own login, see DEPLOY.md § Access control — it takes about a minute.

#The drill companion

The reading lives here; the drilling lives in an interactive page, because flashcards, timers and a sprint tracker work better on a phone than in markdown:

→ Lakehouse Drill Deck

  • Sprint — the ten-day plan as checkable tasks, with progress that persists across devices
  • Cards — 87 flashcards in three yield-sorted tiers, with a "missed only" deck that builds itself from what you actually fumble
  • Drill — whiteboard, diagnostic, design and objection prompts dealt at random against a timer (4 min to draw, 6 to diagnose, 25 to design, 60 seconds to answer an objection)
  • Sheet — the pre-interview cheatsheet

It is private to you unless you share it.

#Map of the content

#00-plan/ — orient before you study

File What it gives you
01-role-and-competency-map.md What is actually being assessed, the competency model, and an honest self-scoring grid
02-interview-loop-playbook.md Stage-by-stage map of both loops, what each interviewer is scoring, how to win each room
03-ten-day-sprint.md The day-by-day plan

#01-platform/ — the technical substance

File What it covers
01-spark-internals.md Catalyst, Tungsten, AQE, DPP, shuffle, joins, memory, Photon, Structured Streaming, Spark Connect
02-delta-lake-internals.md Transaction log, OCC, isolation, data skipping, liquid clustering, deletion vectors, CDF, UniForm/Iceberg
03-lakehouse-architecture-and-topology.md Control/compute plane, workspace and metastore topology, environments, medallion, mesh, DR
04-unity-catalog-governance.md UC object model, privileges, RLS/CLM, ABAC, lineage, system tables, Delta Sharing, federation
05-lakeflow-ingestion-and-pipelines.md Auto Loader, Lakeflow Connect/Declarative Pipelines/Jobs, CDC, streaming patterns, orchestration
06-performance-and-cost.md The hands-on diagnostic playbook — Spark UI triage, skew, spill, MERGE tuning, FinOps
07-ai-and-agents.md Mosaic AI, RAG, Agent Bricks, evaluation, Unity AI Gateway, GenAI reference architectures
08-security-networking-compliance.md Network isolation, Private Link, egress control, CMK, compliance profiles, audit
09-platform-2026-whats-new.md What shipped recently, verified against primary sources — so you are not quoting stale product names
10-competitive-positioning.md Snowflake, Fabric, BigQuery, Flink, real-time stores — honest comparisons and how to talk about them

#02-drills/ — convert knowledge into interview performance

File What it gives you
01-architecture-design-scenarios.md 12 full design scenarios with model answers and the follow-ups you will get
02-deep-dive-qa-bank.md 120+ rapid-fire technical questions with crisp answers
03-customer-objections.md The objection-handling playbook — 25 real objections, scripted
04-behavioral-star-bank.md Values-mapped behavioural questions and a STAR story framework to fill in
05-the-panel-presentation.md How to build and deliver the presentation that decides field loops
06-executive-business-case.md TCO models, board narrative, value framing

#03-practice/ — test yourself under load

File What it gives you
01-mock-loop-full.md A full timed mock loop with scoring rubrics
02-flashcards.md Spaced-repetition deck, sorted by yield
03-whiteboard-prompts.md 30 prompts to draw cold, against a timer
04-self-assessment.md Readiness scoring and gap triage

#99-reference/

File What it gives you
cheatsheet.md The single page to read in the car park before the interview
glossary-and-numbers.md Terminology and the figures worth memorising

#Three rules for this prep

  1. Say it out loud. Every reading block here should end with you explaining the idea to an empty room in under ninety seconds. Reading is recognition; speaking is recall. The interview tests recall.
  2. Draw it. If you cannot draw the architecture from memory in four minutes, you do not know it well enough to draw it while being interrupted.
  3. Depth on demand, not by default. Senior candidates fail by front-loading detail. Lead with the decision and the business consequence; hold the internals until you are asked. Then go three levels deeper than expected.

#A note on freshness

Databricks ships fast, and the platform changed materially at Data + AI Summit in June 2026. 01-platform/09-platform-2026-whats-new.md captures what is current and flags what changed name or status, with sources. Quoting a renamed product is a small error that reads as a big one, so read that file before any interview.

Where a claim in this repo comes from a post-June-2026 announcement, it is cited. Where it is long-standing platform behaviour, it is not. Verify anything you intend to state as a hard number against current documentation on the day — figures and defaults move.