Interview guides · Deepmind

Deepmind Machine Learning Engineer Interview Questions & Guide 2026

Deepmind Machine Learning Engineer interview guide: reported rounds, practice questions with worked solutions and a 7-day prep plan.

The 60-second brief

As a Machine Learning Engineer at DeepMind, you play a pivotal role in advancing the field of artificial intelligence through innovative research and application of machine learning techniques. Your work directly impacts the design and development of systems that can learn from and make decisions based on data. This position is crucial for enhancing DeepMind's products, which span various domains such as healthcare, gaming, and energy efficiency, ultimately aiming to solve some of the world's most pressing challenges.

Ask whether any round happens inside an existing repository instead of a blank file. Reading unfamiliar code, isolating a fault and making the smallest correct change is a different skill from writing a function from scratch, and it needs its own practice.

Focus areas

Detect concurrent edits instead of losing writes
Trace a symptom to a mechanism under load
Paginate large result sets with keyset cursors

Overview

This role is particularly interesting due to the scale and complexity of the problems you will tackle. You will collaborate with cross-functional teams, including researchers and software engineers, to develop cutting-edge algorithms and models that drive state-of-the-art AI technologies. Your contributions will not only enhance user experiences but also drive strategic initiatives that influence the business landscape. Candidates can expect a stimulating environment where creativity meets technical excellence, fostering innovation in the AI field.

The Deepmind interview process

Deepmind candidates report 4 rounds · ≈ 3-5 weeks. The stages below are what candidates describe, not a published process.

Phone Screening

The person on this call usually cannot evaluate your code and does not need to. They write a short paragraph, and that paragraph is what a hiring manager skims when deciding who to put on your loop. So the test is not whether your work was hard, it is whether a non-engineer can repeat it correctly. Name systems by what they did rather than by their internal codename, give each project a shape (what was breaking, what you changed, what happened after), and keep the whole walkthrough near ninety seconds. Depth that cannot survive a paraphrase reads as vagueness.

What to demonstrate

Whether a non-engineer can restate your projects without distorting them, since their paraphrase is what travels to the hiring manager, not your sentences
Whether each project has a shape rather than a stack list: the failure or constraint, the change you made, the result and how it was measured
Whether you can say what was yours inside a team project without either inflating it or disappearing into the plural

How to prepare

Rewrite each headline project as two sentences with no internal system names and no acronyms outside your company, then say them to someone outside engineering and have them repeat them back. Fix whatever came back wrong
Attach one measured number to each project: the baseline, the change, and the window it was measured over. Where nothing was ever measured, say that plainly rather than reaching for a plausible percentage
Time the background walkthrough against a clock. If it runs past two minutes, compress the earliest role to a single clause and spend the recovered time on the most recent one

Technical Rounds

Most of the time lost in this format is not lost to thinking. It goes to a standard-library call you half-remember, an off-by-one in a loop bound, and a debugging loop that mutates code at random until something passes. When output is wrong, stop re-reading the whole function: take the smallest input that reproduces it and walk the state through by hand, printing intermediates if the environment allows. Guessing at a fix without a failing case you understand is how a five-minute bug becomes twenty, and the clock does not pause while you do it.

What to demonstrate

Whether you reach the right structure without a detour, and can write it from memory rather than only recall that one exists
Whether overflow is considered where the language has fixed-width integers, since a signed 32-bit value stops at 2,147,483,647 and then wraps in Java, is undefined behaviour in C++, and does not arise in Python, whose integers grow instead
Whether recursion depth is treated as a constraint on large inputs, given that CPython's default limit is 1000 frames and a deep recursion can exhaust the stack in any language where an iterative version would not
Whether a failing case is isolated and explained before any edit is made to the code

How to prepare

From an empty file and with no references open, implement the pieces you lean on most: a heap push and pop, an iterative DFS with an explicit stack, and a binary search whose midpoint is written lo + (hi - lo) / 2, which avoids the overflow that (lo + hi) / 2 can hit in a fixed-width integer type
Time yourself on the ten library calls you look up most, such as sorting with a custom comparator, splitting and joining strings, and finding the next key at or above a value in an ordered map, until the lookup is gone
Take a solution you know is broken and, before touching it, write one sentence naming the input, the expected value and the actual value. Repeat until you do it without deciding to.

Behavioral Interviews

Your first answer is not really what is scored. It buys the follow-up questions, and those decide the round. An interviewer with fifteen minutes takes one thread and pushes on it four or five times, so a story you can only tell at a single level of detail collapses under the third why. That is an argument for fewer stories known deeply rather than one prepared per prompt. Four or five pieces of work you can still explain down to the code you changed and the argument you had about it will cover nearly anything asked in this round.

What to demonstrate

Whether a story holds as the questioning moves from what you did to why that instead of the alternative, and then to what you would change knowing what you know now
Whether you can re-cut a project to answer the question actually asked rather than delivering a rehearsed block that answers an adjacent one
Whether your level of detail is chosen rather than habitual: going down to the schema when the question is about the data model, staying out of it when the question is about the person who disagreed with you

How to prepare

Pick four projects and write the chain out four levels deep for each: what you did, why that, why not the alternative, and what would have to be true for the alternative to have won. Where you cannot reach the fourth level, you have a placeholder rather than a story
Have someone ask why three times in a row on a single thread with nothing else added, and mark the point where you start repeating a sentence you already said. That point is where the interviewer stops learning anything
Build a one-page index instead of an answer bank: the common prompts in this round (disagreement, a failure that was yours, thin requirements, a deadline you missed, work you inherited) mapped to which of your four projects you would use for each, so the choosing is done now rather than while an interviewer waits

Onsite Interviews

Coding rounds mostly set a floor. They decide whether you clear the bar, not where you land on the ladder. Level tends to come out of the design discussion and the ownership stories, so the question worth auditing beforehand is whether the scope you describe matches the scope of the job. Work that stops at your own service, or a story whose hard part was writing the code rather than getting several people to agree on an interface, reads a level below where you think you are interviewing, and that gap is usually resolved downwards.

What to demonstrate

Whether the largest thing you describe owning ran end to end — the decision, the migration path, the rollout, and what you did when it went wrong — or stopped at the change you merged
Whether design answers include what you would not build, what you would defer, and what you would measure before committing, rather than only what the boxes are
Whether a disagreement in a story was settled with something checkable — a benchmark, a prototype, a written proposal — instead of by seniority or by waiting it out
Whether you can say which calls you made alone and which you escalated, and why the line sat where it did

How to prepare

Write your largest piece of owned work as a timeline of decisions — who decided what, when, and what you did when the plan broke — then delete every sentence whose subject is "we" and see how much survives
Take one system you know well and drill the migration answer: how old and new paths run side by side under live traffic, how you compare their outputs, what the rollback is once writes are going to both, and which step you would not automate
Map each line of the ladder in the job posting to a specific thing you have done, find the line you cannot support, and prepare the closest evidence you have plus an honest account of the gap

Where candidates lose points

Shipping a migration and the code that depends on it as a single change

During any rolling deploy, and for as long as a rollback remains possible, old and new code execute against the same schema at the same time. A migration that drops or renames a column breaks every instance that has not restarted yet, and code that requires a column the migration has not applied breaks every instance that restarted early. The discipline is expand then contract: add the new column nullable, write both shapes, backfill in batches, move reads across once the backfill is verified, and only then stop writing the old shape and drop it - four deploys, usually spread over days. It feels disproportionate until the first rollback, at which point it is the only reason the previous version still runs.

Assuming an isolation level prevents the anomaly you actually have

Isolation levels are named by the SQL standard but implemented differently, so any claim about one is only true of a named engine. PostgreSQL defaults to READ COMMITTED, where every statement takes a fresh snapshot, so two statements inside one transaction can legitimately disagree about the same row. Its REPEATABLE READ is snapshot isolation: it removes non-repeatable and phantom reads but permits write skew, where two transactions each read a set, each conclude their own write is safe, both commit, and the combined result violates a constraint that no single row expresses. Only SERIALIZABLE closes that, and it closes it by aborting a transaction with a serialization failure (SQLSTATE 40001), which means the guarantee is theoretical unless the application has a retry loop. InnoDB's REPEATABLE READ is a different mechanism again - plain SELECTs read a consistent snapshot while locking reads and writes see the latest committed row - so a read-modify-write inside one transaction can act on a value that the transaction's own earlier SELECT never returned.

A queue or buffer with no bound

Every producer-consumer boundary needs a capacity and a policy for reaching it: block the producer, shed load, or drop the oldest entry. Unbounded buffering converts a temporary slowdown into memory exhaustion and hides the backpressure signal that would have revealed the consumer was falling behind.

Issuing one query per row of a result set

Fetch related rows in a single batched query keyed by the ids you already hold, or join them into the original query. A per-row round trip multiplies network latency by the row count, and it looks perfectly fine against the ten rows in your development database.

Questions & practice

Propose a machine learning solution for predicting stock prices.

Medium · machine learning fundamentals

Approach

State the learning problem: the label, the unit of prediction and how the model is used.
Name the simplest model that could work and what would make you move past it.
Say how you would validate it, and where leakage could enter the split.

Follow-up

Where could label leakage enter this setup?
What changes if the classes are heavily imbalanced?

How would you address the challenge of model interpretability in AI?

Medium · machine learning fundamentals

Approach

Say how you would validate it, and where leakage could enter the split.
Pick the metric from the cost of each error type, not from habit.
Name the simplest model that could work and what would make you move past it.

Follow-up

What changes if the classes are heavily imbalanced?
Where could label leakage enter this setup?

Provide an example of a project where you used machine learning to solve a complex problem.

Medium · machine learning fundamentals

Approach

Pick the metric from the cost of each error type, not from habit.
State the learning problem: the label, the unit of prediction and how the model is used.
Say how you would validate it, and where leakage could enter the split.

Follow-up

Where could label leakage enter this setup?
What changes if the classes are heavily imbalanced?

Describe the architecture of a convolutional neural network (CNN).

Medium · machine learning fundamentals

Approach

State the learning problem: the label, the unit of prediction and how the model is used.
Name the simplest model that could work and what would make you move past it.
Say how you would validate it, and where leakage could enter the split.

Follow-up

Where could label leakage enter this setup?
What changes if the classes are heavily imbalanced?

Solve a LeetCode problem related to dynamic programming.

Medium · coding and algorithms

Approach

State the target complexity and say which constraint rules the naive version out.
Name the brute-force solution and its complexity before improving on it.
Choose the data structure from the access pattern, not from familiarity.

Follow-up

Which test case would catch an off-by-one here?
What is the worst case, and how likely is it on real data?

Write a function to perform a depth-first search (DFS) on a binary tree.

Medium · coding and algorithms

Approach

State the target complexity and say which constraint rules the naive version out.
Choose the data structure from the access pattern, not from familiarity.
Walk one small example through your approach before writing the whole thing.

Follow-up

Which test case would catch an off-by-one here?
What is the worst case, and how likely is it on real data?

Diff a projection against the primary without per-row point reads

Hard · reconciliation · range hashing · throttling

The listing projection has drifted and some rows show a stale version. The primary holds 40,000,000 resource rows across 12,000 tenants while serving 1,200 writes and 14,000 reads per second. The obvious repair, reading each resource row and comparing its version against the projection, is correct and would eventually finish. Explain precisely why it is unacceptable here, then give a diff that finds the differing rows, state its complexity, and make it safe to run against a live primary. Replication lag is usually under 100 ms and is not bounded.

Approach

Quantify the naive cost rather than calling it slow: 40,000,000 point reads at even 0.5 ms each is over five hours serialised, and the only lever is concurrency, which is exactly what you cannot spend. The primary's pool is sized for the write path, and 40,000,000 random reads evict the buffer cache that sustains the 85 percent cache hit rate, so the audit degrades the system it is auditing.
Replace random access with one ordered pass per side. Both sides can be read in (tenant_id, resource_id) order, which is a sequential scan on each and a merge join in O(n) time and O(1) memory. For a dense diff that is the whole answer, and it reads the primary once instead of 40,000,000 times.
For the expected sparse case, compare range hashes instead of rows: partition the key space, compute per range an order-independent aggregate over hash(resource_id, version), compare aggregates, and descend only into ranges that differ. With d differing rows and branching factor B, at most d ranges mismatch per level, so the drill-down examines O(d log_B(n/d)) ranges and reads full rows only in mismatching leaves.
Aggregate with a sum modulo 2^64 or a multiset hash, never XOR. XOR is order-independent but self-cancelling, so two rows wrong in the same way, or a row duplicated on one side, leave the range aggregate matching and the range is declared clean.
Pin the comparison to a point in time or it reports lag as drift: consider only rows whose updated_at is older than now minus a lag margin, and re-check each candidate mismatch individually before repairing. At 1,200 writes per second a diff without this reports thousands of false positives, and an unattended repairer would then overwrite live rows with stale values.
Make the run resumable and throttled: batch by range key, persist the last completed range, and watch a signal such as replica lag or primary CPU, pausing rather than pressing on. A reconciliation that cannot be stopped and resumed gets killed halfway and restarted from zero, which is how a repair becomes an incident.

Worked solution (35 min)

Compute the naive cost explicitly at 40,000,000 reads and 0.5 ms each, then at 100 concurrent, and state what those connections do to a pool already carrying 1,200 writes per second.
Write the merge-join version over (tenant_id, resource_id) and state its memory.
Define the range aggregate: the range key, the per-row hash input, and the combining function, with one sentence excluding XOR.
Work an example with 40,000,000 rows, branching factor 256 and 5 differing rows, and count the ranges examined.
Add the watermark filter and the resume point, and name the throttle signal the loop watches.

Expected result: A rejection of per-row point reads backed by the time cost and the cache-eviction argument, a single ordered merge join as the dense-case answer at O(n) time and O(1) memory, a range-hash drill-down examining O(d log_B(n/d)) ranges using a sum or multiset hash rather than XOR, a watermark excluding recently written rows, and a resumable throttled run loop.

Verify it yourself

Injecting one known stale row makes exactly one leaf range mismatch and the drill-down reaches it.
Duplicating a row on one side is detected, which is precisely the case XOR would miss.
Running the diff twice against an idle system reports zero mismatches both times.

Follow-up

The diff reports 900 stale rows. How do you decide between patching those rows and rebuilding the projection from resource_revision?
Same job, but the projection lives in a search index that cannot be scanned in key order. What changes?
How would you run this continuously at low cost instead of only as incident response?

Stop tag and share joins from fanning out a page

Medium · join fan-out · semi-joins · many-to-many

resource_tag is (resource_id, tag_id) with PK (resource_id, tag_id); resource_share is (resource_id, shared_with_user_id, permission). The tagged-and-shared listing inner-joins resource to both, filters tenant_id, tag_id = ANY($2) and shared_with_user_id = $3, orders by updated_at DESC and takes 50. Pages come back with fewer than 50 distinct resources and the total in the header is far too high. Explain the row multiplication, rewrite both the page query and the count query so each is correct, and name the index each one needs. PostgreSQL 16.

Approach

Do the arithmetic against the predicates that are actually there. An inner join emits one row per matching child row, and both joins are filtered: tag_id = ANY($2) admits only the requested tags, shared_with_user_id = $3 admits one user's share rows. So a resource holding three of the requested tags and shared with $3 once yields three rows, not one — the multiplier is its count of matching tags times its share rows for that single user, and that second factor is 1 unless the table admits duplicate (resource_id, shared_with_user_id) pairs. LIMIT 50 then limits rows rather than resources, and COUNT(*) counts pairs — the header is the product, not the population.
Reject DISTINCT as the fix. It deduplicates after the product has been built, so the planner must materialise and sort the fanned-out set before the LIMIT can apply, and it leaves any SUM or AVG in the same select list wrong.
Rewrite both filters as semi-joins, keeping resource as the only row source: AND EXISTS (SELECT 1 FROM resource_tag rt WHERE rt.resource_id = r.resource_id AND rt.tag_id = ANY($2)) and the same shape against resource_share. A semi-join stops at the first match per resource and preserves the driving index order, so ORDER BY updated_at DESC, resource_id DESC LIMIT 50 still stops after 50 rows.
Count with the same predicates and no join at all: SELECT count(*) FROM resource r WHERE r.tenant_id = $1 AND r.status = 'active' AND EXISTS (...) AND EXISTS (...). Nothing multiplies a resource, so the number is the population.
Attach the tags for display after the page has been cut — LEFT JOIN LATERAL (SELECT array_agg(rt.tag_id) FROM resource_tag rt WHERE rt.resource_id = p.resource_id) ON TRUE over the 50 returned rows. Aggregate over the page, never over the tenant.
Index both directions and say which query each serves: PK (resource_id, tag_id) serves the lateral lookup, (tag_id, resource_id) serves the EXISTS probe by tag, and resource_share needs (shared_with_user_id, resource_id) for the same reason. An index covering one direction only leaves the other as a scan.

Worked solution (30 min)

Build a tenant where each resource carries 0-5 tags from a 20-tag vocabulary and is shared with 0-4 distinct users, then bind $2 to three tags and $3 to a user holding shares on about half the resources. Run the joined query and compare its row count to the distinct resource count on page one.
Run COUNT(*) on the joined shape and on the EXISTS shape and compare both to a ground truth computed from distinct ids; then give $3 a second permission row on 10% of resources and record which of the two counts moves.
EXPLAIN both page queries and compare rows-read plus the presence of a Sort or HashAggregate node above the join.
Add (tag_id, resource_id), re-run the EXISTS probe, and record the plan change on the inner side.

Expected result: The EXISTS page returns exactly 50 distinct resources and reads about 50 driving rows with no aggregate above the join. The EXISTS count matches the ground truth. The joined COUNT(*) is larger by roughly the average number of matching tags per qualifying resource — the tags it holds from $2, not its total tag count — times its share rows for $3, which is 1 until the duplicate permission rows exist and rises with them afterwards. The EXISTS count is unchanged by those duplicates.

Verify it yourself

The page's resource_id list has no duplicates and is 50 long whenever at least 50 rows match.
count(*) from the rewrite equals count(DISTINCT r.resource_id) over the joined shape.
EXPLAIN shows a semi-join node on the inner side rather than a full join followed by a unique or aggregate step.

Follow-up

The filter changes from 'any of these tags' to 'all of these tags'. Rewrite it and state what it costs relative to the ANY form.
A resource can be shared with the same user twice under different permissions. Does your count change, and should it?
Where does the correct total come from when the tenant holds 4M resources and the header must not cost 200 ms?

Find version gaps and relay lag with window functions

Medium · window functions · event tables · frame semantics

outbox_event holds event_id, aggregate_type, aggregate_id, aggregate_version, event_type, payload, status ('pending','published','dead'), attempts, created_at, published_at. A projection is missing rows and you must decide whether the relay skipped events or the consumer dropped them. Write three queries over the last seven days: one listing every aggregate_id whose published aggregate_version sequence has a hole, one giving per-day counts with a running total, and one returning the newest published event per aggregate. For each, say where the window function is evaluated relative to WHERE and LIMIT. PostgreSQL 16.

Approach

Gaps: compute lead(aggregate_version) OVER (PARTITION BY aggregate_id ORDER BY aggregate_version) in a subquery, then filter next_version <> aggregate_version + 1 in the outer query. Window functions are evaluated after WHERE, GROUP BY and HAVING and before the outer ORDER BY and LIMIT, so the predicate cannot sit in the same WHERE clause and PostgreSQL 16 has no QUALIFY.
Say what the seven-day filter does to the answer: it truncates every partition, so the first row per aggregate has no predecessor inside the window and a hole spanning the boundary is invisible. Widen the window, or join to resource.version as the authority for the true maximum.
Running total: SELECT date_trunc('day', created_at) AS d, count() AS n, sum(count()) OVER (ORDER BY date_trunc('day', created_at) ROWS UNBOUNDED PRECEDING). An aggregate inside a window call is legal because grouping runs before windowing. The grouping key is unique per row here so ROWS and RANGE agree, but write the frame anyway — over ungrouped rows with tied timestamps the default RANGE frame pulls in every peer row and the total jumps.
Newest per aggregate: DISTINCT ON (aggregate_id) ... ORDER BY aggregate_id, aggregate_version DESC is the cheap PostgreSQL-only form when an index matches that order; row_number() OVER (PARTITION BY aggregate_id ORDER BY aggregate_version DESC) = 1 is the portable form and needs a subquery for the same evaluation-order reason as the gap query.
Interpret rather than report: no gaps plus a normal p95 of published_at - created_at points at the consumer; gaps or a fat lag tail point at the relay; rows still 'pending' with attempts > 0 point at neither, because they never left the database.
Be explicit that the partial index on (created_at, event_id) WHERE status = 'pending' does not serve any of these — they read published rows. Name the index a recurring monitor would need, and say why a query run twice a year may not deserve one.

Follow-up

Relay failover redelivers events. Does a duplicate break the gap query, and how would you detect one from this table alone?
Turn the gap check into a continuous monitor rather than a query someone runs after an incident. What does it watch?
The consumer claims it never received event 4,812,006. What do you look at, in what order?

What strategies would you use to monitor and maintain model performance over time?

Medium · ml system design and production

Approach

Fix the product goal and the online metric before choosing any model.
Name what you would monitor after launch and what triggers a retrain.
Separate the offline training path from the online serving path.

Follow-up

How would you detect drift before the metric drops?
How would you roll the new model out safely?

How would you design a recommendation system for a large-scale e-commerce platform?

Medium · ml system design and production

Approach

Name what you would monitor after launch and what triggers a retrain.
Separate the offline training path from the online serving path.
Say where features come from at serving time and how they match training.

Follow-up

How would you roll the new model out safely?
How would you detect drift before the metric drops?

What considerations would you take into account for data privacy in your models?

Medium · system design

Approach

Name the failure you are designing for, then the recovery path.
Name the read and write paths separately; they rarely have the same bottleneck.
Fix the scope first: who calls this, how often, and what they do when it fails.

Follow-up

How does this behave when that dependency is down for an hour?
What would you drop to keep the system up under load?

What is the difference between supervised and unsupervised learning?

Medium

Approach

Clarify what is being asked and what a complete answer contains.
State your assumptions explicitly before working the problem.
Work from the requirement backwards to the design.

Follow-up

How would you know your answer was wrong?
What assumption would you test first?

Rebuild the search projection while it serves nine thousand queries

Medium · projections · eventual consistency · rebuild · watermark

The read-model service answers about 9k queries/second at a 120 ms p99 from a projection built off the event log, normally under 2 seconds behind. A mapping change forces a full rebuild from resource_revision, during which apply lag rises to minutes. Writes continue at 1.2k/second and events at 4k/second. Design the rebuild: how the new index is populated and cut over, how position is tracked per log partition, what the API returns alongside results so a client can tell a stale answer from a current one, and the criterion for cutting over.

Approach

Build into a second index and swap an alias rather than mutating the live one. The rebuild is then reversible by pointing the alias back, so a bad mapping costs a wasted rebuild instead of an outage. The price is peak storage for two full copies and double apply load during catch-up, and both numbers should be stated up front rather than discovered when disk fills.
Track position per log partition, not globally. Order is guaranteed only within an aggregate's partition, so progress is a vector, and the only number safe to publish is taken from the least advanced partition - the slowest one is what bounds completeness. Publishing the most advanced partition's position declares the projection current while another partition sits twenty minutes behind.
Make apply idempotent so the backfill and the live tail can overlap without a freeze. Each document records the aggregate_version it reflects, and any event at or below that version is discarded. This is why the event carries the full fact and not a delta: a delta cannot be discarded safely, and a consumer that calls back to read current state applies a state newer than the event it is processing, which is how a projection ends up with changes applied out of order.
Turn lag into a contract instead of a surprise. Return the watermark with every result set, and return the version produced by a write so the client can compare the two. A client that wrote version 7 and receives results at a watermark older than its own commit can show that its change is still landing, rather than rendering the previous value as current. Blocking the read until the projection catches up would convert a staleness problem into an availability problem at 9k queries/second, and choosing not to do that is the trade.
Define the cutover numerically. Writes are untouched by the rebuild - they commit to the primary and land in the outbox - so the only coupling is apply throughput. If catch-up applies slower than the 4k events/second arriving, it never converges. The cutover criterion is that lag is measurably decreasing and below a stated threshold, not that the backfill loop reached the end of its range.

Worked solution (30 min)

Write the rebuild sequence: snapshot source, populate the second index, attach the live tail, verify, swap alias, retire the old index - naming what is reversible at each step.
Define the per-partition position record and the single watermark derived from it, and show what each would report when one partition stalls.
Write the idempotent apply rule using the document's recorded aggregate_version, and replay one event twice against it.
State the cutover threshold in lag terms and the measurement that shows convergence rather than completion.

Expected result: A parallel index populated from resource_revision snapshots plus the live tail, cut over by an alias swap that is reversible. Position is stored per log partition and the published watermark comes from the least advanced one. Apply is idempotent on the document's recorded aggregate_version, so backfill and tail can overlap without a freeze. Results carry the watermark and writes return their version so clients can detect staleness. Cutover happens when lag is decreasing and under the stated threshold.

Verify it yourself

Applying the same event twice leaves the document unchanged.
The published watermark never advances past the least advanced partition.
Pointing the alias back restores previous behaviour with no reindex.
Catch-up rate exceeds the 4k events/second arrival rate before cutover, shown by the lag trend rather than by the loop finishing.

Follow-up

The rebuilt index disagrees with the primary tables for 300 documents. Which is authoritative, and how do you decide without freezing writes?
The rebuild doubles load on the log and pushes the projection p99 from 120 ms to 400 ms. What do you throttle, and which signal sets how much?
Clients start polling until the watermark passes their write. What does that do at 9k queries/second, and what do you offer instead?

p99 jumped on one listing filter while p50 stayed flat

Medium · query planning · composite index · percentiles · tail latency

After a release that added an owner_user_id filter to the resource listing, p99 rose from 90 ms to 1.9 s while p50 stayed at 40 ms. Traffic and row counts are unchanged. resource carries the index (tenant_id, status, updated_at DESC, resource_id DESC). The new query filters tenant_id and owner_user_id, orders by updated_at DESC, resource_id DESC, and takes 20 rows. On PostgreSQL, explain the shape of the regression, prove it from a query plan, and give the index you would add.

Approach

Start from the shape. A flat p50 with a moved p99 means a subset of requests changed cost, not all of them, so the first job is naming the subset. Bucket the endpoint's latency by the tenant's row count; the natural hypothesis is that large tenants are a small share of requests and all of the tail.
Get the plan for the new query on a large tenant with EXPLAIN (ANALYZE, BUFFERS). Expect an index scan over the tenant's range, a filter discarding most of it, then a Sort feeding the Limit, possibly reporting Sort Method: external merge Disk. Read actual rows on the scan node, not estimated.
Explain why the existing index cannot serve it. A composite B-tree is seekable only as a left prefix, and with no equality predicate on status the scan cannot treat updated_at as an ordering, because rows in the tenant's range are ordered by status first. Everything matching must be read and sorted before LIMIT 20 can apply, so a tenant with 400,000 rows pays 400,000 rows to return 20.
Add (tenant_id, owner_user_id, updated_at DESC, resource_id DESC). Equality on the first two columns leaves the index ordered by updated_at within that pair, so the plan becomes an index scan that stops after 20 rows with no Sort node. PostgreSQL can scan a B-tree backwards, so the DESC markers matter only if the two sort columns ever disagree in direction; keeping them explicit documents the order the keyset cursor depends on.
Price the fix. This is a fourth index on a table taking about 1.2k writes/second, and every insert and version bump maintains it. Justify it against the query it serves, and check whether it makes an existing index redundant, which here it does not, since the original still serves the status-filtered default listing.
Re-measure per tenant-size bucket rather than in aggregate. A fleet-wide p99 can improve while the largest tenant is still on the old plan.

Follow-up

The endpoint paginates with OFFSET. What does page 500 cost with your index, and what does the keyset version cost?
How would you have caught this before release, given that a 10,000-row seed database produces the same plan shape at an unnoticeable cost?
If a fourth index were unacceptable on write grounds, what else could serve this query?

Suggested practice checklist

Day one measures instead of guessing, under a fixed rubric, and the remaining hours are allocated in proportion to the gaps before any studying begins. The allocation is deliberately not renegotiated midweek, because the area that feels worst on day three is usually the one that is moving.

Day 1 — Diagnostic, scored before you study anything

Sit a 110-minute diagnostic in four blocks: forty-five minutes on two coding problems, twenty-five on one design prompt taken to interface and data model, twenty of short-answer fundamentals, and twenty delivering two behavioural answers aloud.
Score each block from 0 to 3 on a fixed rubric where 3 is correct and fluent, 2 is correct but slow or prompted, 1 is partially correct and 0 is stuck, grading the artifact rather than how the attempt felt.
Allocate days two to five in proportion to 3 minus each block's score, write the allocation down, and commit to leaving it alone.

Deliverable: A scored rubric and a fixed hour allocation for the rest of the week.

Day 2 — Largest gap: find the boundary rather than the subject

Split the weakest area into named sub-skills and rate each separately. For coding those are restating the problem, choosing the structure, stating the invariant, turning the invariant into loop bounds, handling empty and single-element input, and accounting for complexity out loud.
Attempt three items positioned just above where the rating drops off, and for each write the first move you failed to make.
Re-attempt one of them from blank four hours later with nothing open.

Deliverable: A sub-skill map with the two blocking sub-skills circled.

Day 3 — Drill the blocking sub-skill by repeating the shape

Do eight short repetitions of the same shape rather than eight different problems, so what gets practised is the pattern and not the puzzle.
State the rule you now hold in one sentence, then test it against a case built to break it, a sliding window over an array containing negative values, or a cache-aside read path whose invalidation message is dropped.
Have someone else read your one-sentence rule and find the precondition you left out.

Deliverable: One rule statement with its preconditions attached and one counterexample that would have caught the incomplete version.

Day 4 — Second gap, plus maintenance on the strongest area

Run the same sub-skill decomposition on the second-largest gap in half the time.
Spend twenty-five timed minutes on the block you scored highest, choosing the hardest item you can still finish rather than a warm-up.
Write whether each area fails you on recall, on setup, or on execution, and set the fix accordingly: repetition for recall, a written checklist for setup, timed work for execution.

Deliverable: A second sub-skill map plus a one-line failure diagnosis for each area.

Day 5 — The gap that is not a skill

Record one technical and one behavioural answer, then count two things in the playback: seconds before your first clarifying question, and sentences you began without knowing where they would end.
Practise saying that you do not know, followed by how you would find out, without letting it soften into a guess, and practise stating a complexity or an estimate before being asked for it.
Redeliver one answer under a hard ninety-second cap, which forces structure ahead of detail.

Deliverable: Two recordings with a counted reduction in time-to-first-question.

Day 6 — Retest under day-one conditions

Sit the same 110-minute structure with new prompts of comparable difficulty and score it on the identical rubric.
For any block that did not move, change the method rather than adding hours: a block stuck at 1 usually means the practice was too varied, not too short.
Write down which single block you would still lose the offer on.

Deliverable: A second scored rubric placed beside the first, with one named remaining risk.

Day 7 — Full loop under interview conditions

Run a sixty-minute mock over the two blocks that moved least, with an interviewer briefed to interrupt and change direction mid-answer.
Write the recovery script for going blank: restate the question, state your assumption, name the first thing you would check.
Say every rule from the week aloud without reading it, and cut any you cannot state in a single sentence, since a rule you have to reconstruct mid-answer will not survive an interruption.

Deliverable: A one-page card holding the recovery script and only the rules you could state from memory.

Know where your numbers came from

Bring the two or three numbers the story rests on and know how they were collected. A p99 whose timer starts inside your handler excludes the time a request spent queued, so it can sit flat while users wait longer. Give the window, the percentile and what the measurement left out, or drop the number.

Describe a time when you faced a significant challenge in a project.

Medium · behavioural and collaboration

Approach

State the situation in two sentences and spend the rest on the reasoning.
Give the blast radius: what could have broken, and what you measured.
Name the disagreement and how you resolved it with evidence.

Follow-up

How did you know your change caused the improvement?
What would you do differently if you ran that again?

Narrate an outage you owned from page to postmortem

Hard · incident response · blast radius · postmortems

Pick an incident you personally drove, ideally one where writes were affected rather than reads. In six to eight minutes: state the symptom as it first appeared on a dashboard, the blast radius you established before you knew the cause, the mitigation you applied and when, the mechanism you eventually proved, and the follow-up that would prevent a repeat. Bring numbers: error rate, tenants affected, minutes to mitigate, minutes to resolve. If you cannot name what you measured, choose a different incident.

Approach

Open on the signal rather than the cause: which metric at which percentile moved, on which service, at what time, so the listener follows the same evidence you had rather than a conclusion you already reached.
Separate mitigation from diagnosis out loud. State what you did to stop the bleeding (flag off, shed traffic, drain a lease, roll back a deploy) and say plainly that you did it before the mechanism was known, because those are two jobs with different deadlines.
Establish blast radius in countable terms: how many tenants, how many writes, and crucially whether the effect was loss or only delay. An append-only revision table or a pending outbox row means the change survived and the projection was merely behind, which is a repair rather than a data-loss incident.
Prove the mechanism instead of asserting it. Name the trace span that grew, the plan that flipped to a sequential scan, the lease that expired, plus one alternative you ruled out and the signal that stayed flat while you ruled it out.
Close on the durable fix and its cost, distinguishing what landed that week from what needed an expand-and-contract migration across several deploys, and say which of the two you actually finished.

Follow-up

What would you do differently in the first five minutes, given the same dashboard and no more information?
Which follow-up action did you deliberately not take, and why was dropping it the right call?
How did you convince yourself the mitigation was safe to apply while the cause was still unknown?

Argue against a design, lose, and commit anyway

Medium · disagreement · service boundaries · decision records

Describe a design you argued against and lost. State the failure you predicted as a named mechanism, not a feeling about complexity: two services that would need one transaction, a projection with no rebuild path, a write path with no idempotency key. Say what evidence you brought, what the decision maker weighed instead, and what you did after the decision was made: what you instrumented, what you wrote down, and whether the prediction came true. Five minutes.

Approach

State the prediction in falsifiable form up front: the mechanism, the condition that triggers it, and the observable outcome. A prediction that cannot be checked also cannot be credited to you later.
Show the evidence you had at the time and label each piece honestly as measured, analogous, or intuition. Keeping the intuition is fine; disguising it as data is the thing that erodes your standing in the next argument.
Represent the opposing case at full strength, including the constraint you did not control: a fixed date, a team boundary, or the fact that the decision was cheap to reverse and yours was not.
Make disagree-and-commit concrete. Name the artefact you left behind so the prediction could be settled without you: the alert and its threshold, the counter on the dashboard, the decision note that recorded the trade-off and the condition that would revisit it.
Report the outcome without editing it. If the design held and your predicted mechanism never fired, say so and say what you had mis-weighted, which is more persuasive than a vindication story.

Follow-up

What threshold on that alert would have proved you right, and did anyone ever look at it?
If the same proposal arrived tomorrow with the same deadline, would you argue it the same way?
How did you behave toward the design once it shipped and started failing in a different way than you predicted?