DE 100: data engineering interview questions

Write a SQL query to find the time period when the most people were online, measured in seconds.

MediumCodingAmazon

Turn each session into a +1 event at login and a -1 event at logout, take a running SUM over the events in time order, and report the second or interval where that running count peaks.

Solve it in Coding Practice: When the most people were online

Given a large table with 3 columns (datetime, employee, customer_response), find the top 10 employees with the most phone numbers found in the customer_response column.

HardCodingAmazon

Pull the phone numbers out of the free text with a regular expression, count the matches per employee, and keep the top 10 with ORDER BY and LIMIT, stating how ties break.

Solve it in Coding Practice: Employees whose call notes hold the most phone numbers

Calculate first day retention rate - players who log in the 2nd day immediately after first login.

MediumCodingAmazon

Find each player's first login date with MIN, join the logins on the day right after it, and divide the players who came back that day by all players.

Solve it in Coding Practice: First-day retention of new players

Write a SQL query to get total revenue generated by each subscriber in the year 2014.

MediumCodingAmazon

Filter the orders to 2014 with a date range on the raw column, join them to the subscribers, and SUM the revenue per subscriber with GROUP BY.

Solve it in Coding Practice: Revenue per subscriber in 2014

Calculate the median number of searches a person made last year using summary table.

HardCodingGoogle

The summary table pairs each search count with how many people made it, so take a running SUM of those people and pick the count where the running share crosses one half.

Solve it in Coding Practice: Median searches per person from a summary table

Calculate the sum of odd-numbered and even-numbered measurements separately for a particular day.

MediumCodingGoogle

Number the day's measurements in time order with ROW_NUMBER, then SUM with CASE WHEN on whether the row number is odd or even, both totals in one pass.

Solve it in Coding Practice: Odd and even measurement sums per day

Write a SQL query that returns the 2nd-highest salary in the engineering department.

EasyCodingNetflix

Rank the department's salaries with DENSE_RANK in descending order and keep rank 2, so a tie at the top does not hide the answer; return NULL when there is no second salary.

Solve it in Coding Practice: Second-highest salary in Engineering

Write a query to find customers who placed more than three orders in the past year.

MediumCodingMicrosoft

Filter the orders to the last twelve months, GROUP BY customer, and keep the groups with HAVING COUNT(*) > 3.

Solve it in Coding Practice: Customers with more than three orders in the past year

Write SQL query to find ranks of employee salaries, ordered by department.

MediumCodingMicrosoft

Use RANK or DENSE_RANK OVER (PARTITION BY department ORDER BY salary DESC), and say which one you chose and how it treats ties.

Solve it in Coding Practice: Salary ranks within each department

Write a query to return all neighborhoods that have 0 users (LEFT JOIN with IS NULL).

MediumCodingMicrosoft

LEFT JOIN neighborhoods to users and keep the rows where the user key IS NULL; NOT EXISTS expresses the same anti join.

Solve it in Coding Practice: Territories with no employees

Calculate average vacant days across AirBnbs in 2021 for active properties only.

HardCodingAirbnb

For each active listing, count the days of 2021 not covered by a booking, clipping bookings to the year, then average those days across the listings.

Solve it in Coding Practice: Average vacant days in 2021 for active listings

Get average rating of each Airbnb property listing per month.

MediumCodingAirbnb

Truncate the review date to its month, then AVG the rating grouped by listing and month.

Solve it in Coding Practice: Average rating per listing per month

Find top 10 listings with most bookings by joining hosts and bookings tables.

MediumCodingAirbnb

Join bookings to listings and hosts, COUNT the bookings per listing, ORDER BY that count descending and LIMIT 10, with a rule for ties.

Solve it in Coding Practice: Top 10 listings by bookings, with their host

Write a query to get list of customers with their last purchase details.

MediumCodingGoogle

Number each customer's purchases with ROW_NUMBER OVER (PARTITION BY customer ORDER BY purchase time DESC) and keep row 1, which returns the whole purchase row.

Solve it in Coding Practice: Each customer's last purchase

How do you optimize a slow SQL query? Explain query tuning process.

HardTheoryGoogle

Measure first: read the execution plan to find the costliest step, then read less data, help the engine find rows, fix joins and statistics, and check the result is unchanged.

Write a function to find non duplicate numbers preserving order: [1,1,3,2,5,6,5] → [1,3,2,5,6]

MediumCodingAmazon

Walk the list once with a set of values already seen and keep each value the first time it appears; list(dict.fromkeys(items)) does the same in one line.

Solve it in Coding Practice: Remove duplicates, keep the order

Given a json object with nested objects, write a function that flattens all objects to single key-value dictionary.

HardCodingAmazon

Recurse through the object, joining parent and child keys with a separator such as a dot, and decide up front how lists are handled.

Solve it in Coding Practice: Flatten a nested JSON object

Write code to find maximum combinations of infinite coins {1,2,5} that add up to 20 rupees.

HardCodingAmazon

Count with dynamic programming: ways[0] = 1, then for each coin add ways[amount - coin] to ways[amount] up to the target; looping over the coins first counts combinations, not orderings.

Solve it in Coding Practice: Ways to make an amount from coins

Extract unique values from a dictionary where values occur only once.

MediumCodingUber

Count the values with collections.Counter and keep those whose count is 1, in the order they first appear.

Solve it in Coding Practice: Dictionary values that occur only once

What's the difference between list, tuple, and set in Python for data pipelines?

EasyTheory

A list for ordered batches you append to, a tuple for fixed records and composite keys, a set for fast membership checks and de-duplication; at millions of rows, move to pandas, Arrow or Spark.

Explain Python modules used in data engineering: Pandas, NumPy, Requests, SQLAlchemy.

EasyTheoryGoogle

NumPy for fast arrays, pandas for tabular cleaning and joins, Requests for API calls with timeouts and retries, SQLAlchemy for pooled, parameterized database access; show how they combine in one small pipeline.

How would you optimize a data pipeline written in Python using vectorized operations?

MediumTheory

Profile to find the slow loop, replace row logic with column operations such as np.where, .str, .dt, merge and groupby().transform, use compact types, and check the output is unchanged.

Write Python code to rank hosts based on number of beds using pandas groupby.

MediumCodingAirbnb

Total the beds per host with groupby("host_id")["beds"].sum(), then rank with rank(method="dense", ascending=False) or a sort, and state how ties rank.

Solve it in Coding Practice: Rank hosts by their number of beds

What are generators in Python and how are they useful in large data processing?

MediumTheoryNetflix

A generator yields one value at a time, so memory stays flat however large the input; chain generators into lazy pipelines, and remember each can be consumed only once.

How do you handle large CSV files that don't fit in memory using Python?

MediumTheorySpotify

Never load it whole: stream the rows, read with chunksize, load only the columns and types you need, convert it once to Parquet, or use DuckDB, Polars or Spark.

Explain the Global Interpreter Lock (GIL) and its impact on data processing.

HardTheory

The GIL lets one thread run Python bytecode at a time, so threads help I/O-bound work but not CPU-bound Python; use processes, vectorized libraries or Spark for heavy computation.

Write a Python class for a data pipeline with error handling and logging.

HardCodingStripe

Give the class extract, transform and load methods and a run method that calls them in order, log each step, catch specific exceptions with context, and re-raise failures so the caller sees them.

Solve it in Coding Practice: A pipeline class with error handling and logging

How do you implement multithreading vs multiprocessing for data processing in Python?

HardTheoryUber

Threads for I/O-bound work, processes for CPU-bound Python, both through concurrent.futures; send small inputs to processes, bound the workers and handle failures per task.

Write code to detect longest patterns in song plays from user listening data.

HardCodingSpotify

Sort each user's plays by time and walk them, counting consecutive plays of the same song and keeping the longest run; in SQL it is the gaps and islands pattern.

Solve it in Coding Practice: Longest repeat streak in song plays

How do you handle missing data and data cleaning in pandas efficiently?

MediumTheoryMeta

Measure the missing values per column and agree what missing means, then drop, fill or flag column by column; coerce types explicitly, remove duplicates and validate before writing.

Design a relational database schema for a ride-sharing app like Uber.

HardDesignAmazon

Model riders, drivers, vehicles, rides, ride events, payments and ratings, treat the ride status as a state machine, keep location pings out of the core tables, and analyse a warehouse copy.

Design a database schema for an e-commerce platform with products, orders, and customers.

MediumDesignAmazon

Normalize customers, addresses, products, prices, orders, order lines, payments and shipments; copy the unit price onto each order line, store money as DECIMAL, and build a star schema for analytics.

Present a design of a gaming company database with players, games, and achievements.

MediumDesignMeta

Players, games and achievements, with link tables from players to games and from players to achievements; derive leaderboards, partition sessions by date, and model facts for sessions and unlocks.

Design a fact table for retail sales database with appropriate measures and dimensions.

MediumDesign

Fix the grain first, one row per transaction line, then add the dimension keys, the degenerate keys and additive measures; store the parts of a ratio, not the ratio, and partition by date.

Explain difference between Star Schema and Snowflake Schema with examples.

MediumTheory

Both put a fact table in the centre; a star keeps each dimension as one denormalized table, a snowflake normalizes dimensions into related tables, trading fewer joins against less redundancy.

Design a dimension table for customer entity with at least five attributes.

EasyDesign

A surrogate key, the natural key, descriptive and geographic attributes, SCD Type 2 columns for history, audit columns, and an unknown member so no fact is dropped.

How do you handle slowly changing dimensions (SCDs) Type 1, 2, and 3?

HardTheoryMicrosoft

Type 1 overwrites, Type 2 adds a row with validity dates and a new surrogate key, Type 3 adds a column for the previous value; choose per attribute and test late and same-day changes.

Design a schema to keep track of customer address changes over time.

MediumDesignAmazon

Keep an address history table with valid_from and valid_to, one current row per customer and address type, no overlapping periods, and orders that copy the address they used.

What are factless fact tables and when would you use them?

MediumTheory

A fact table with dimension keys and no measures, counted rather than summed: event tables for attendance or logins, coverage tables for finding what did not happen.

Explain normalization vs denormalization trade-offs in data modeling.

MediumTheoryGoogle

Normalization stores each fact once for safe, fast writes in OLTP; denormalization repeats data for simple, fast reads in analytics, built from the normalized source.

Design a data model for tracking user behavior on a streaming platform like Netflix.

HardDesignNetflix

Land raw client events, build playback session, impression, search and rating facts with profile, content and device dimensions, and handle late events, retries and privacy.

How do you model many-to-many relationships in dimensional modeling?

MediumTheory

Put a bridge table between the fact and the dimension, with a weighting factor or group keys, so totals are not counted twice.

Design ERD for a social media platform with users, posts, and interactions.

HardDesignMeta

Users, follows, posts, media, likes, comments, hashtags, mentions and notifications, with keys that block duplicates, soft deletes, counters updated in the background and a precomputed feed.

What is dimensional modeling according to Ralph Kimball methodology?

MediumTheory

Choose the business process, declare the grain, identify the dimensions, then identify the facts; around them sit conformed dimensions, the bus matrix, surrogate keys and SCDs.

Design a data model for IoT sensor data with time-series considerations.

HardDesign

Separate device and sensor reference data from narrow, append-only readings partitioned by time; store event time and ingest time, downsample, compress, and flag gaps instead of inventing values.

Design an end-to-end ETL pipeline that ingests payment data into analytics warehouse.

HardDesignStripe

Change data capture and settlement files into an immutable raw zone, staged and modelled payment facts, reconciliation with the provider, idempotent MERGE loads, quality gates and tight security.

How would you design a data pipeline for ingesting data with increasing volume?

MediumDesignAmazon

Buffer the ingestion, partition on an even key, process incrementally and idempotently, use scalable table formats and elastic compute, and alert on growth before it hits a limit.

Design data pipeline for hourly user analytics with batch and streaming options.

HardDesignUber

Land the events once, then offer an hourly batch job that overwrites its partition and a streaming job with event-time windows and watermarks; in a hybrid, batch corrects the stream.

How do you handle data quality issues and validation in data pipelines?

MediumTheoryNetflix

Test completeness, uniqueness, validity, consistency and timeliness at ingestion, after transforms and before publishing; quarantine bad records, block bad publishes and fix causes upstream.

How would you manage incremental data loads in a data warehouse?

MediumTheory

Detect changes with CDC, a watermark or an increasing id, advance the watermark only after the commit, MERGE idempotently, expect late data and validate every run.

When should you consider streaming over batch processing?

MediumTheoryNetflix

Stream when data loses its value within seconds and something will act on it, as in fraud checks or alerts; batch is simpler and cheaper for hourly or daily consumers.

Explain the key stages in ETL process: Extract, Transform, Load.

EasyTheory

Extract with little load on the source, transform by cleaning, validating, conforming and modelling, load idempotently with bulk operations, and check counts and totals.

How do you implement error handling and recovery mechanisms in ETL processes?

MediumTheory

Classify errors: retry the transient ones, quarantine bad records, stop on fatal ones; make steps idempotent, checkpoint progress, keep the raw data and write runbooks.

What are the differences between ETL and ELT approaches?

MediumTheory

ETL transforms on a separate engine before loading; ELT loads raw data first and transforms it inside the warehouse; compare compute, flexibility, privacy and cost.

How do you handle schema changes in source systems during ETL?

MediumTheoryMicrosoft

Check the schema on every run, handle each kind of change (new, removed, renamed, retyped, changed meaning) on its own terms, and keep a raw layer and explicit column lists.

Design a real-time data pipeline to process payment transactions and update accounts.

HardDesignStripe

An idempotent payment write with an outbox to Kafka partitioned by account, a double-entry ledger that skips redelivered events, dead letter handling and daily reconciliation.

How do you optimize slow-running ETL jobs for better performance?

MediumTheoryAmazon

Find the bottleneck first, then process less data, use columnar formats, push work to the data, fix joins and parallelism, and replace row-by-row logic.

How do you implement data lineage tracking in ETL pipelines?

HardTheory

Capture lineage at run time with OpenLineage, from SQL with dbt or a parser, and from the platform; record datasets, jobs and versions, and use lineage in change reviews and alerts.

Design ETL process to collect and process tickets in real-time manner.

HardDesignAmazon

Collect ticket events into a stream, process them with validation, enrichment, routing and service level timers, then serve agent dashboards, alerts and warehouse facts.

How do you handle late-arriving data in streaming ETL pipelines?

HardTheoryNetflix

Process by event time with watermarks, choose the lateness bound from measured delays, emit updated results, capture events that are too late, and let batch correct the numbers.

Design a scalable data processing system for Netflix's recommendation engine.

HardDesignNetflix

An offline layer trains candidate and ranking models from the lake, a feature store keeps features the same in training and serving, and an online service ranks precomputed candidates with fresh context.

Design a system similar to BookMyShow handling concurrency and fault tolerance.

HardDesignUber

Hold seats with one atomic conditional update and an expiry, take payment with an idempotency key, confirm in one transaction, and reconcile payments with bookings when something fails.

Design a data warehouse to capture and analyze sales data.

MediumDesignAmazon

Gather the business questions, land the sources incrementally, build a Kimball star schema on conformed dimensions, add tests and metric definitions, and start with one business process.

Design a real-time analytics platform for a ride-sharing company like Uber.

HardDesignUber

Stream ride and location events through Kafka into per-zone, per-minute windows, serve them from a real-time OLAP store, and keep every event in the lake for batch correction.

Design distributed storage system for petabyte-scale data.

HardDesignGoogle

Split files into large blocks, replicate or erasure-code them across racks, run a replicated metadata service, place data with consistent hashing, and add a table format for ACID tables.

Design monitoring and alerting system for data pipelines.

MediumDesign

Monitor job health, freshness, data quality, streaming lag and cost, and route each alert with its impact, links and a runbook to the owning team, without alert fatigue.

Design data lake architecture for storing multi-structured data.

HardDesignAmazon

Raw, cleaned and curated zones on object storage with a table format, a catalogue entry for every unstructured file, and governance, compute and operations across the zones.

Design API for high-frequency trading data with low latency requirements.

HardDesign

Two paths: a push-based binary feed with sequence numbers and in-memory order books for trading systems, and a historical query API over a time-series store for analysts.

Design caching strategy for frequently accessed data in distributed system.

MediumDesign

Cache data that is read often and changes rarely, usually cache-aside with TTLs; invalidate on writes, and plan for stampedes, missing keys, hot keys and cache outages.

Design backup and disaster recovery strategy for critical data systems.

MediumDesign

Agree RPO and RTO per system, follow the 3-2-1 backup rule with point-in-time recovery, choose a recovery architecture by cost, and test restores regularly.

Design A/B testing platform for data-driven product decisions.

HardDesignMeta

Configure experiments, assign users by deterministic hashing, log exposures, compute metrics in one shared pipeline, and use sound statistics with guardrails and a sample ratio check.

Design real-time fraud detection system for payment processing.

HardDesignStripe

Score each payment synchronously from real-time and batch features, combine rules with a model and thresholds, learn from chargebacks, and fall back to rules on a timeout.

Design data platform for processing IoT sensor data at scale.

HardDesign

Devices publish over MQTT to a gateway and a stream, stream processing validates and alerts, and storage splits hot time-series data from the lake and long-term rollups.

Design search and recommendation system for e-commerce platform.

HardDesignAmazon

Search retrieves keyword and vector candidates and ranks them with learning to rank; recommendations precompute candidates in batch and rank them at request time.

Design data governance framework for enterprise data management.

MediumDesign

Set out roles, policies, processes, tools and measures, start with the most critical domains, and make data easier to find and trust, not only harder to reach.

What is PySpark and how does it differ from Apache Spark?

EasyTheory

Spark is the distributed engine on the JVM and PySpark is its Python API, so keep heavy work in DataFrame operations and built-in functions that run inside the engine.

Explain concept of RDD in PySpark and how to create one.

MediumTheory

An RDD is an immutable, partitioned collection that rebuilds lost partitions from its lineage; create one with parallelize, textFile or df.rdd, and prefer DataFrames today.

What is SparkSession and how do you create one in PySpark?

EasyTheory

SparkSession is the single entry point since Spark 2.0: build it with SparkSession.builder and getOrCreate(), and let spark-submit set the master on a cluster.

Explain concept of lazy evaluation in PySpark with examples.

MediumTheory

Transformations only build a plan and actions run it, which lets Catalyst optimize the whole plan; cache a DataFrame that several actions reuse.

What are broadcast variables in PySpark and when to use them?

MediumTheory

A broadcast variable ships a read-only value to each executor once; a broadcast join applies the same idea to a small table, as long as it fits in memory.

How do you optimize PySpark jobs for better performance?

HardTheoryNetflix

Start from the Spark UI and the plan, then read less, prefer built-in functions to Python UDFs, optimize joins and shuffles, handle skew and right-size the resources.

What is role of Catalyst optimizer in PySpark?

MediumTheory

Catalyst resolves the plan, rewrites it with rules such as predicate pushdown and column pruning, picks a physical plan using size and cost, and generates code; adaptive execution re-plans at run time.

Write PySpark code to perform groupBy operation and calculate average of column.

MediumCoding

Group and aggregate with df.groupBy("key").agg(F.avg("value")), using built-in functions so the work stays inside the engine.

Solve it in Coding Practice: Average track length per genre

What are UDFs in PySpark? Write example and demonstrate usage.

MediumCoding

A UDF wraps a Python function so Spark can apply it to a column: declare it with F.udf and a return type, use it in withColumn, and prefer a built-in or a pandas UDF when one exists.

Solve it in Coding Practice: A UDF that formats track length

Explain difference between DataFrame and RDD in PySpark.

MediumTheory

An RDD holds opaque objects transformed by your functions; a DataFrame has a schema that Catalyst and Tungsten optimize, so it is usually much faster in PySpark.

Write PySpark code to join two DataFrames and explain different join types.

MediumCoding

Call df1.join(df2, on="key", how=...), and explain inner, left, right, full outer, left semi and left anti joins, with a broadcast when one side is small.

Solve it in Coding Practice: Join artists to albums, keeping every artist

What is PySpark Streaming and how to stream data using TCP/IP protocol?

HardTheory

Structured Streaming treats a stream as an unbounded table; the socket source is for learning only because it cannot replay data, so production reads from Kafka or files.

Explain the architecture of Hadoop and its core components (HDFS, MapReduce, YARN).

MediumTheoryNetflix

HDFS stores replicated blocks under a NameNode, YARN allocates the cluster's resources, and MapReduce processes data in map, shuffle and reduce phases.

How do you handle data skew in Spark applications?

HardTheoryUber

Find the hot keys, then rely on adaptive skew handling, broadcast the small side, separate null keys, salt the hot keys or process them on their own path.

Write PySpark code to perform window functions for calculating running totals.

HardCoding

Define a window partitioned by customer, ordered by date, with a ROWS frame from unbounded preceding to the current row, and take F.sum("amount").over(window).

Solve it in Coding Practice: Running total of each customer's spend

What are OLTP vs OLAP systems and their impact on database design?

MediumTheory

OLTP runs many small, consistent transactions on normalized row stores; OLAP runs large scans on denormalized, columnar models; keep them apart and move data with ETL or ELT.

What is materialized view in BigQuery and when to use it?

MediumTheoryGoogle

A stored query result that BigQuery refreshes automatically and incrementally; use it for repeated aggregations that must stay fresh, after checking its SQL limits.

Explain concept of data lake vs data warehouse differences.

MediumTheory

A warehouse holds modelled data with schema on write for BI; a lake holds raw data of any type with schema on read; a lakehouse adds table formats to the lake.

What are components of data warehouse architecture (staging, integration, access)?

MediumTheory

Staging lands raw copies, integration cleans, conforms and keeps history as the single source of truth, and access shapes marts and metric definitions for the people who use them.

How do you handle schema evolution in data warehouse environment?

MediumTheoryMicrosoft

Prefer additive changes, stage breaking ones through a deprecation period, version and review schema changes, agree contracts with sources, and keep raw data for backfills.

What are best practices for data warehouse performance optimization?

MediumTheory

Model for the queries, read less data through partitioning, clustering and column pruning, write efficient SQL, load incrementally, and tune the slowest queries first.

Explain concept of data mart vs data warehouse and when to use each.

EasyTheory

A warehouse integrates data for the whole organization; a mart serves one subject or team, best built from the warehouse on conformed dimensions.

What are different types of indexes in data warehouse and their use cases?

MediumTheory

B-tree indexes for selective lookups, bitmap indexes for low-cardinality columns, sort keys and zone maps for skipping blocks, and Bloom filters for point lookups.

How do you implement data retention policies in warehouse environment?

MediumTheory

Set retention per class of data with legal and the business, partition by date so expiry drops partitions, automate the enforcement, archive where required and cover every copy.

What are partitioning strategies for large tables in data warehouse?

HardTheory

Range partitioning by date is the usual choice, with list, hash and composite options when queries need them; avoid very high-cardinality keys and cluster on them instead.

Tell me about the most challenging data engineering project you worked on.

MediumBehavioralAmazon

Answer with a STAR story: situation, task, action, result. What the interviewer listens for: the size and difficulty of the problem, your own part in it, the technical decisions you made, and what you learned.

Describe a time you had to optimize a data pipeline for performance. What was your approach?

MediumBehavioralNetflix

Answer with a STAR story: situation, task, action, result. What the interviewer listens for: a method (measure, find the cause, fix, verify) rather than a lucky guess.

How do you prioritize multiple competing data engineering tasks and deadlines?

MediumBehavioralAmazon

Answer with a STAR story: situation, task, action, result. What the interviewer listens for: judgment about impact and urgency, and clear communication when something has to wait.

Tell me about a time you applied judgment to a data decision when complete data was not available.

MediumBehavioralAmazon

Answer with a STAR story: situation, task, action, result. What the interviewer listens for: sound reasoning under uncertainty, assumptions stated out loud, and a plan to check them.

How would you convey complex data insights to non-technical stakeholders?

MediumBehavioralUber

Answer with a STAR story: situation, task, action, result. What the interviewer listens for: that you lead with the answer, choose plain words and one clear visual, and connect the insight to a decision.

Describe a situation where you had to collaborate with cross-functional teams on a data project.

MediumBehavioralMeta

Answer with a STAR story: situation, task, action, result. What the interviewer listens for: how you align different goals, agree definitions and handle disagreement.

Tell me about a time you made a mistake in a data project and how you handled it.

MediumBehavioralAmazon

Answer with a STAR story: situation, task, action, result. What the interviewer listens for: ownership, speed and honesty in fixing it, and prevention afterwards.

How do you stay updated with the latest data engineering technologies and trends?

EasyBehavioral

Answer with a STAR story: situation, task, action, result. What the interviewer listens for: a real, sustainable habit, and evidence that learning turns into better work.

Describe a time you had to learn a new technology quickly for a project.

MediumBehavioralSpotify

Answer with a STAR story: situation, task, action, result. What the interviewer listens for: a learning plan under time pressure, and delivery.

How do you ensure data quality and reliability in your engineering projects?

MediumBehavioralStripe

Answer with a STAR story: situation, task, action, result. What the interviewer listens for: concrete practices (tests, monitoring, ownership) shown on a real project.