Company interview guide
DoorDash Machine Learning Interview Guide (2026)
Timed SQL and marketplace case studies for data scientists; project deep dives, coding and infra-heavy ML design for ML engineers.
DoorDash's machine learning interview loop has 7 reported stages, from recruiter call or written questionnaire to partner, values and hiring-manager rounds; typical length: 2-4 weeks.
Roles covered: Data Scientist and ML Engineer. The question bank holds 10 questions reported from DoorDash interviews, 2 of them free to solve in your browser.
- Interview process length
- 2-4 weeks
- Stages
- 7
- Reports linked to bank questions
- 10
- Bank report dates
- Feb 2022–Feb 2026
- Questions in bank
- 10 (2 free)
- Last reviewed
- October 2026
How the DoorDash interview loop works
A recruiter call or an emailed questionnaire of about 15 written questions, then a technical screen, then a virtual final loop. Data science: a 60-minute screen split into 30 minutes of SQL and a 30-minute case, then three case rounds of about 45 minutes plus a partner or hiring-manager conversation. ML engineering: a project deep dive and a 60-minute coding round, often back to back, then an onsite with coding, ML system design, an ML domain round and a hiring-manager behavioral. Rejections after the screen are often quick and come without feedback.
The DoorDash interview process, stage by stage
01Recruiter call or written questionnaire
About 30 min call; some candidates instead answer roughly 15 open-ended questions by email
A short check of background, motivation and logistics. Several candidates skipped the call entirely and were sent a written questionnaire, and referred candidates sometimes went straight to the technical screen.
What it tests
- Why DoorDash and why this team
- How your experience maps to the role and level
- Salary expectations, location and timeline
How to prepare
- Treat the written questionnaire like an interview: specific, concise answers that name your strongest relevant project.
- Ask the recruiter for the exact format and length of the screen, since several candidates were surprised by a different format on the day.
- Read the prep material the recruiter sends; data science candidates get a guide that describes the screen closely.
Recruiter or HR call in 20 data science and 3 ML reports; 4 reports got an emailed list of about 15 written questions instead, and 3 had no call at all.
02Data science: SQL and case screen
60 min over video: 30 min SQL (four questions of rising difficulty in a shared editor, often with no way to run code) plus a 30 min product case, usually with two interviewers
The most consistently reported round at DoorDash. The SQL half asks four questions on order, delivery or customer tables, and many candidates do not finish the last one. The case half is a short, vague marketplace problem where you set metrics, form hypotheses and propose a test. A few teams used a take-home or a hiring-manager call instead of, or before, this screen.
What it tests
- Joins, CTEs, aggregation, CASE and date filters under time pressure
- Window functions: RANK, LAG and month-over-month comparisons
- Reading an existing query and explaining or changing it
- Structuring a vague marketplace problem: clarify, hypothesize, pick metrics
- Proposing an experiment and naming guardrails
How to prepare
- Practise four SQL questions in 30 minutes in a plain editor without running anything; speed and correct logic matter more than narration here.
- Expect to retype or rewrite earlier CTEs for later questions; do not rely on reusing your previous answer.
- For the case, always walk through consumers, merchants and Dashers before recommending anything.
- Know DoorDash's core marketplace health metrics before the call: order volume, delivery time, cancellations, Dasher supply.
SQL plus case screen in 50 of 60 data science reports (41 since 2024), usually 30 + 30 minutes with four SQL questions; the company's round 1 prep guide describes this format.
03Data science: final loop case studies
Virtual onsite: usually 3 case rounds of about 45 min each, sometimes one led by the hiring manager
Longer versions of the screen case, with more follow-up questions and less guidance. Prompts are DoorDash scenarios such as a new Dasher vehicle type, a delivery-fee coupon, a wider delivery radius or a drop in orders in one city. Candidates report heavy probing on experiment design, sizing and trade-offs between the three sides of the marketplace. One 2025 loop instead ran statistics, ML and coding rounds after a take-home.
What it tests
- A/B test design: unit of randomization, minimum detectable effect, duration
- Causal reasoning when a clean experiment is not possible
- Diagnosing a metric change by segmenting across customers, merchants and Dashers
- Opportunity sizing and lifetime value with stated assumptions
- Trade-offs such as cannibalization and network effects
How to prepare
- Prepare a reusable structure (clarify, hypotheses, metrics, test, decision) but adapt it to each prompt; interviewers notice canned answers.
- Be ready for experiment follow-ups: why not stop when the sample size is reached after two days, urban versus rural effects, interference between sides.
- Study how DoorDash makes money and its flywheel; several candidates were asked about it directly.
- Say what you would do with no experiment possible; at least one round per loop tends to be deliberately ambiguous.
Two to four case rounds in 14 data science final-loop reports, 13 since 2024; one 2025 loop used statistics, ML and coding after a take-home. The official prep guide names a final round but not its format.
04ML engineering: project deep dive
About 45 min on one past ML project, often back to back with the coding round; some teams share a question document beforehand
The usual first technical round for ML engineers. You pick one project and the interviewer goes deep on the problem, the design choices, why you chose a metric or model over the alternatives, and the outcome. One 2025 candidate was downleveled after this round for lacking online deployment experience, so production details matter.
What it tests
- Framing: the business problem and how success was measured
- Why this model, these features and this metric rather than the alternatives
- How the model was deployed, served and monitored
- What went wrong and what you would change
How to prepare
- Choose a project you owned end to end, including online serving, and lead the conversation rather than waiting for questions.
- If you receive a question document beforehand, fill it in; one candidate found it made the round much easier.
- Prepare numbers: data size, latency, offline and online metric changes.
Project deep dive in 10 of 14 ML-role reports from 2024-2026, usually about 45 minutes and paired with a coding round in the same session.
05ML engineering: coding
45-60 min live coding, one or two problems, in the screen and again onsite; DoorDash says engineering coding rounds are moving to a 60 min AI-assisted session
Reported problems are mostly medium algorithm questions, sometimes wrapped in a delivery scenario: ordering tasks from prerequisites, a maximum over interval-value pairs, closest value in a BST, tree and simulation tasks. One ML scientist screen asked for k-means from scratch. In 2026 the company announced that engineering coding rounds become a working session in your own IDE with AI assistants allowed; it is not yet confirmed which ML loops use it, and one 2026 ML candidate still reported a medium-hard live coding round.
What it tests
- Graphs and topological ordering
- Intervals, trees and simulation with clean state handling
- Turning a delivery-flavored story into a standard algorithm
- For ML scientist screens, implementing a classic ML algorithm in numpy
How to prepare
- Ask your recruiter whether your coding round is the classic format or the AI-assisted working session, and set up your IDE and assistant if it is the latter.
- For the AI-assisted format, narrate as you go and verify generated code with tests; orientation, debugging and scope control are scored.
- Drill topological sort and interval problems; they show up more than once in reports.
Live coding in 10 ML-role reports, mostly one medium problem per hour; the company's engineering blog (2026) describes replacing classic coding rounds with an AI-assisted session.
06ML engineering: ML system design and ML domain
Onsite: one infra-leaning ML system design round (some loops had two) plus an ML domain round; one design question came up in a screen
Design rounds lean toward infrastructure: serving a recommendation feed at low latency, caching and scaling, or an ML platform with a feature store and monitoring. A separate ML domain round asks for a modeling approach to a DoorDash problem, such as targeting promotions for consumer growth on a budget. Evidence here is thin, so expect variation by team.
What it tests
- Online serving for ranking or recommendations: caching, fallbacks, latency budgets
- Feature stores that serve both training and real-time inference
- Reproducibility, versioning and monitoring for drift and degradation
- Modeling incremental lift for promotions under a budget
- Choosing offline and online metrics for a marketplace model
How to prepare
- Prepare to go deep on infrastructure, not only model choice; one candidate was told DoorDash ML engineers own both.
- Reuse DoorDash examples: delivery ETA, dispatch, store ranking, fraud and promotions.
- For promotions, talk about uplift rather than raw conversion and how you would measure it with a holdout.
Free reported questions for this stage
- Serving Infrastructure for a Delivery-App Recommendation FeedML System DesignHardSolve
ML system design or ML domain rounds in 4 ML-role reports from 2025-2026; public write-ups describe the same pair of an infra-leaning design round and an ML domain round.
07Partner, values and hiring-manager rounds
Usually one 30-45 min conversation with a cross-functional partner (often a PM) plus hiring-manager time, which may include a case
Data science loops include a cross-functional or values conversation, and the hiring manager often adds behavioral questions or another case. ML engineering loops end with a hiring-manager behavioral that can turn into a practical problem, such as how to stop wrong items reaching customers.
What it tests
- Conflict with a manager or stakeholder, and how it resolved
- Working with product and operations partners
- Ownership and impact of past projects
- Practical judgment on operational problems, not only ML ones
How to prepare
- Prepare several STAR stories on conflict, prioritization and influence; one candidate said the partner round asked the most basic versions.
- If a hiring manager poses an operational problem, consider simple process fixes before proposing a model.
Free reported questions for this stage
- Preventing Wrong-Item Deliveries: Diagnosis, Pickup Check and RolloutProduct & MetricsMediumSolve
Partner, values or hiring-manager behavioral round in 15 data science and 2 ML-role final-loop reports; the company's careers site names an engineering values chat only for early-career software engineers.
Questions and themes reported most often
- Timed SQL with window functions (rank, lag, month-over-month) on order and delivery tables (16 reports, data science: sql and case screen)
- Project deep dive: why this metric, this model, this design (10 reports, ml engineering: project deep dive)
- Diagnose a marketplace metric change: order drop, slower deliveries, cancellations, cold food (9 reports, data science: final loop case studies)
- Reason across consumers, merchants and Dashers in every case (9 reports, data science: final loop case studies)
- Design and size an A/B test for a feature, fee or promotion change (8 reports, data science: final loop case studies)
- Success metrics for a new product or feature launch (7 reports, data science: sql and case screen)
- Should Dashers deliver by bicycle or scooter, and how would you measure it (5 reports, data science: final loop case studies)
- Order tasks from a list of prerequisites (topological sort) (2 reports, ml engineering: coding)
What DoorDash asks about
- Product cases & experimentation38%
- SQL20%
- Coding (algorithms)13%
- ML system design & domain12%
- Project deep dive9%
- Behavioral & partner8%
DoorDash interview prep checklist
- Solved four SQL questions with CTEs and window functions in 30 minutes without running code
- Ran three 45-minute mock marketplace cases with heavy follow-up questions
- Can size an experiment and defend its duration, unit and guardrails
- Can explain DoorDash's business model and flywheel in two minutes
- ML engineering: rehearsed a 45-minute deep dive on one project, including deployment
- ML engineering: sketched a low-latency recommendation serving design with a feature store and fallbacks
- Prepared conflict and stakeholder stories for the partner round
Tips
- Data science: the one-hour SQL plus case screen is the main filter, so rehearse it end to end with a timer and no code execution.
- Learn the three-sided marketplace well enough to reason about how any change hits consumers, merchants and Dashers.
- Know experiment design deeply: randomization unit, minimum detectable effect, duration, interference and what to do when you cannot run a test.
- Expect terse interviewers and little feedback; many candidates who felt they did well were still rejected, so do not over-read signals in the room.
- Do not use AI tools or search during live interviews unless the interviewer invites it; the company says this can disqualify you.
- ML engineering: pick one production project to defend in depth, including serving and monitoring, and prepare to lead that conversation.
- Confirm with your recruiter whether coding is classic live coding or the newer AI-assisted session, and practise the matching format.
- Read DoorDash engineering posts on ETA, dispatch, ranking and feature infrastructure before ML design rounds.
See the rest of the DoorDash loop
The later stages, the questions reported for them and how to prepare are in Pro.
Unlock with ProPractice the DoorDash loop
Work the stages in order: every reported question for this company, grouped by round, with solutions in the app.
DoorDash interview FAQ
How many interview rounds does DoorDash have for ML roles?
Candidates report 7 stages: Recruiter call or written questionnaire, Data science: SQL and case screen, Data science: final loop case studies, ML engineering: project deep dive, ML engineering: coding, ML engineering: ML system design and ML domain and Partner, values and hiring-manager rounds. A recruiter call or an emailed questionnaire of about 15 written questions, then a technical screen, then a virtual final loop. Data science: a 60-minute screen split into 30 minutes of SQL and a 30-minute case, then three case rounds of about 45 minutes plus a partner or hiring-manager conversation. ML engineering: a project deep dive and a 60-minute coding round, often back to back, then an onsite with coding, ML system design, an ML domain round and a hiring-manager behavioral. Rejections after the screen are often quick and come without feedback.
How long does the DoorDash ML interview process take?
Reported timelines run 2-4 weeks (2-4 weeks, longer when interviews are rescheduled).
What does DoorDash ask in ML interviews?
By reported question mix: Product cases & experimentation 38%; SQL 20%; Coding (algorithms) 13%; ML system design & domain 12%; Project deep dive 9%; Behavioral & partner 8%. The most repeated themes: Timed SQL with window functions (rank, lag, month-over-month) on order and delivery tables; Project deep dive: why this metric, this model, this design; Diagnose a marketplace metric change: order drop, slower deliveries, cancellations, cold food.
How should I prepare for the DoorDash ML interview?
Data science: the one-hour SQL plus case screen is the main filter, so rehearse it end to end with a timer and no code execution. Learn the three-sided marketplace well enough to reason about how any change hits consumers, merchants and Dashers. Know experiment design deeply: randomization unit, minimum detectable effect, duration, interference and what to do when you cannot run a test.
Are there free DoorDash interview questions?
Yes. LeetML has 10 questions reported from DoorDash interviews, and 2 of them are free to solve in the browser; they are listed by stage above.
Related interview guides
- UberGrid and graph coding, ML from scratch and marketplace system design built around ETA, pricing and Uber Eats.
- InstacartModerate coding, deep questions on classical ML and your own models, and grocery-flavoured ML design and experiment cases.
- LyftProbability and marketplace cases for Data Scientists; coding, ML framing and ML design for ML Engineers.
- AirbnbMarketplace ranking design, a hard look at your past models, one practical coding round and a values check.
- GrabAn online test with ML, SQL and coding, then technical, case-study and manager rounds rooted in ride-hailing, delivery and payments.
- ShopifyPractical pair programming with AI allowed, a signature Life Story conversation, and commerce-flavoured ML design and modeling.