Lakehouse Prepchief architect

#The Interview Loop Playbook

How each round works, what the interviewer is actually scoring, and how to win it.

#Part 1 — The Databricks / field-engineering loop

Publicly reported structure for Field Engineering roles (Solutions Architect, Specialist SA, and the senior/principal/chief variants). Exact composition varies by region and level; treat this as the shape, not a guarantee.

# Round Length Who What they score
1 Recruiter screen 30 min Recruiter Motivation, level fit, comp alignment, communication
2 Hiring manager 60 min HM Experience depth, territory/engagement fit, values, "would I put this person in front of my biggest customer"
3 Technical screen / deep dive 60 min Senior SA / architect Real platform depth, hands-on credibility, honesty at the edge of knowledge
4 Technical assignment take-home — Structure, clarity, correctness (not all roles/levels)
5 Panel presentation 60 min 3–6 people role-playing a customer The decisive round. Discovery, narrative, depth under pushback, objection handling, coachability
6 Cross-functional panel 2–3 × 60 min Peers, partner teams, leadership Collaboration, values, exec presence, breadth
7 References → offer — — —

#The rounds in detail

Round 2 — Hiring manager. This is a fit-and-narrative round, not a trivia round. They want a clean story of your career arc and evidence you operate at the level. Have ready: your two-minute arc, your three strongest engagements, why this company and this role now, and one question that shows you understand their business problem rather than their product.

Round 3 — Technical deep dive. Expect to be taken to the bottom of at least one topic. Typical shapes:

  • "Walk me through what happens when I run this query / this job."
  • "A customer says their job got 3× slower after a data volume increase. Go."
  • "Explain the Delta transaction log. Now: two writers, same table, same second — what happens?"
  • "When would you not use Databricks?"

The scoring is as much about how you handle the edge of your knowledge as the knowledge itself. The correct move when you do not know: "I don't know that specifically. Here's how I'd reason about it, and here's how I'd verify." Then actually reason well. Bluffing is detected instantly and is usually terminal — the value "we are truth seeking" makes this explicit.

Round 5 — The panel presentation. See ../02-drills/05-the-panel-presentation.md for the full build. In outline: you get a scenario in advance, you present to a panel playing customer stakeholders, and they interrupt. Widely described by candidates as the round that decides field loops. They are scoring:

  • Did you ask questions before proposing? (Most-cited failure.)
  • Did you pitch to the business outcome or to the feature list?
  • Did you hold up under interruption without losing the thread or getting defensive?
  • Could you go deeper when pushed and shallower when the "CFO" spoke?
  • Did you take feedback and adjust live?

A known hazard: panels sometimes role-play disengaged or hostile stakeholders and will not react to prompts. This is a test, not rudeness. Do not let flat affect in the room derail you — name it gracefully ("I'll keep going and leave time at the end for the questions I'd normally draw out here") and continue with energy.

Round 6 — Cross-functional and values. Databricks publishes six cultural principles and interviews against them close to one-to-one:

We are customer obsessed · We raise the bar · We are truth seeking · We operate from first principles · We bias for action · We put the company first

Map one prepared STAR story to each. See ../02-drills/04-behavioral-star-bank.md. Culture fit here is weighted, not ceremonial.

#Part 2 — The enterprise chief architect loop

Customer-side loops are shaped differently: less demo, more governance and money.

# Round What they score
1 Recruiter / talent partner Level, comp, motivation
2 Hiring manager (CDO / CTO / Head of Architecture) Arc, scope of prior ownership, opinion quality
3 Target-state architecture exercise Can you design the estate, not just a pipeline
4 Technical deep dive with principal engineers Are you still real, or a slide architect
5 Business case / exec panel TCO, value narrative, board-readiness
6 Stakeholder / cross-functional (security, finance, product, data science) Influence without authority; do peers want you
7 Sometimes: a written architecture submission Structure and written clarity

#What differs from the field loop

#Part 3 — Answer frameworks

Four frameworks. Internalise them so you never freeze on structure and can spend your attention on content.

#Framework 1 — CLAIM for design questions

Context → Levers → Architecture → Implications → Measure

  • Context (ask, don't assume): volumes, latency SLO, users and skills, regulatory posture, incumbent estate, budget, timeline. 3–5 questions, then summarise back what you heard.
  • Levers: name the 2–3 decisions that actually determine the outcome. This is where seniority shows.
  • Architecture: draw it. Ingest → store → transform → serve → govern → observe.
  • Implications: what you gave up, what breaks at 10×, what you would revisit in six months.
  • Measure: how you will know it worked — cost per query, freshness, SLA, adoption.

Most candidates do A. Seniors do C, L, I and M. L and I are where the score is.

#Framework 2 — EVIDENCE for debugging questions

Symptom → Scope → Evidence → Hypothesis → Test → Fix → Prevent

Never propose a fix before naming the evidence you would look at. Full playbook in ../01-platform/06-performance-and-cost.md.

#Framework 3 — STAR-L for behavioural questions

Situation → Task → Action → Result → Learning. The L is what separates senior from mid. Keep S and T to two sentences; most airtime on A; R has a number; L is honest.

#Framework 4 — FEEL-FELT-FOUND-FACT for objections

"I understand why that's a concern (feel) — a lot of teams have hit that (felt) — what we've generally found is X (found) — and specifically, here's the mechanism/number (fact)."

The fact is mandatory. Without it this is salesmanship; with it, it is architecture. Full scripts in ../02-drills/03-customer-objections.md.

#Part 4 — Room mechanics

Opening. Ninety seconds, rehearsed, never more. Arc → current scope → why this role. Then stop talking and ask them what they most want to cover. This buys you the agenda.

When interrupted. Stop immediately, answer the question asked, then explicitly return: "— coming back to where I was…". Never finish your sentence over an interviewer. Candidates who plough on score badly on collaboration.

When you don't know. State it in one sentence, reason from first principles, offer how you would verify. Then move on with confidence — do not apologise twice.

When you're wrong. "You're right, I had that backwards — it's actually X." Fast, clean, no spiralling. This scores positively against truth-seeking. Candidates who defend a wrong answer for three minutes lose more than the point.

Time discipline. In the presentation, land the core message by 60% of the slot. Assume you will lose a third of your time to interruptions — that is the design, not an accident.

Closing. Have three questions that only someone who did the work would ask. See ../02-drills/06-executive-business-case.md §7.

#Part 5 — Red flags interviewers write down

Memorise this list. It is the scorecard's free-text box.

  • "Did not ask a single question before designing."
  • "Listed features, could not justify a choice."
  • "Could not go deeper when pushed — knowledge was one layer thick."
  • "Product knowledge was out of date."
  • "Defensive when challenged."
  • "Every example was a success; no evidence of learning."
  • "Could not explain the business value in non-technical terms."
  • "Claimed credit ambiguously — unclear what they did versus the team."
  • "No numbers in any example."
  • "Talked for eleven minutes without a breath."

The inverse of each of these is a preparation task.