2023–2026 DoorDash Machine Learning Engineer Intern Interviews: 9 Experiences
- Company
- DoorDash
- Reports
- 9 interview reports
- Level
- Internship
- Stages
- Onsite
In the newest loop, a PhD internship interviewed in late 2026, the coding hour was an AI-assisted data task: turn a handful of messy order tables into a labeled training set.
Coverage: DoorDash machine learning internships, interviews from late 2023 to late 2026: eight Machine Learning Engineer Intern loops (six from PhD candidates and two from Master's candidates, including Summer 2024 and Summer 2026 openings) and a Fall 2025 Machine Learning Intern co-op in San Francisco that required a Master's degree. Stages: an online assessment in one loop (not described), phone screens, virtual onsites of two rounds, and team-match calls in two loops.
Full-time reports are kept apart: DoorDash ML interviews for experienced hires and DoorDash ML interviews where the hiring level wasn't reported.
What came up
Every one of the nine loops had a coding round. In seven of them it sat next to some kind of ML conversation: knowledge questions, an open case about features and metrics, or a recommendation or search design. Behavioral questions showed up in five loops, though none of them had a round billed as behavioral only.
| Topic | Reports | Most recent report |
|---|---|---|
| A coding round | 9 of 9 | Nov 7, 2026 (report date) |
| Behavioral or engineering-values questions | 5 of 9 | Nov 7, 2026 (report date) |
| Coding with an AI assistant | 1 of 9 | Nov 7, 2026 (report date) |
| ML knowledge questions | 4 of 9 | Feb 2026 (interview date) |
| ML case studies | 4 of 9 | Feb 2026 (interview date) |
| Recommendation and personalization models | 4 of 9 | Feb 2026 (interview date) |
| Past projects and résumé | 4 of 9 | Feb 2026 (interview date) |
| Team-match call | 2 of 9 | Feb 2026 (interview date) |
| Coding on time ranges or time points | 2 of 9 | Late 2025 (interviews; reported Feb 7, 2026) |
| Coding on a grid | 2 of 9 | Oct–Dec 2023 (interviews; reported Feb 3, 2024) |
Questions about LLM experience and LLM research interests came up once (04); the knowledge table below lists them next to the named ML questions, while 06's basic ML questions aren't named, so they have no row there. A report can appear in more than one row.
Questions we found
Four coding rounds aren't named: a medium-level problem in a DoorDash setting (02), a string-and-dictionary task (03), one the candidate solved with recursion and dynamic programming (05), and one known only by its topics, trees and simulation (06). Also not described: 01's online assessment, the case studies in 02 and 03, the basic ML questions in 06, the behavioral prompts in 01 and 02, the other questions asked after 04's coding, and all of 07's behavioral prompts but one. Everything else is listed below, including questions with no FastPrep practice version.
Algorithms / Data Structures
| Question | Frequency | All last seen dates |
|---|---|---|
| Fewest batches for a list of time points, with at most B points per batch and every point within W of the batch's first time (practice) | 1 report | Feb 7, 2026 (report date; interviews late 2025) |
| Follow-up: Meeting Rooms II (LeetCode 253) recast as riders taking orders (practice) | 1 report | Feb 7, 2026 (report date; interviews late 2025) |
| Most coupon credit usable at one moment, then the time ranges where it happens (practice) | 1 report | Reports: Feb 3, 2024 (report date; interview Dec 2023); Additional FastPrep history: Feb 19, 2025 |
| Longest Increasing Path in a Matrix (LeetCode 329), then a second algorithm (practice) | 1 report | Feb 3, 2024 (report date; interviews Oct–Dec 2023) |
| On a 0/1 grid of restaurants, the starting cell with the smallest Manhattan distance to all of them (practice) | 1 report | Jan 24, 2024 (report date; interviews Oct–Dec 2023) |
AI Coding
| Question | Frequency | All last seen dates |
|---|---|---|
| Build a labeled training dataset from five small order and store tables, handling duplicates and missing values, with an AI assistant (practice) | 1 report | Nov 7, 2026 (report date) |
ML System Design
| Question | Frequency | All last seen dates |
|---|---|---|
| Personalized restaurant recommendations for a user's search query, then where an LLM fits (practice) | 1 report | Reports: Feb 7, 2026 (report date; interviews late 2025); Additional FastPrep history: Nov 7, 2026; Mar 20, 2026 |
| A recommendation system: one for food delivery, or a recommender's features, model and evaluation metrics (practice) | 2 reports | Reports: Sep 1, 2025 (report date); Jan 24, 2024 (report date); Additional FastPrep history: Mar 22, 2026; Mar 20, 2026; Dec 19, 2025; May 13, 2025; Apr 5, 2025; Jan 9, 2025; Aug 9, 2024; Jul 21, 2024; Jul 19, 2024; Jul 4, 2024; Jun 23, 2024; Oct 24, 2023; Oct 12, 2023; Mar 8, 2023; Sep 20, 2022; May 16, 2022; Feb 11, 2021 |
| A case set in a business context the candidate didn't name: feature treatment, and how you'd judge whether a new model performs well | 1 report | Feb 3, 2024 (report date; interviews Oct–Dec 2023) |
| Fraud detection: features, model and evaluation metrics (practice) | 1 report | Reports: Jan 24, 2024 (report date); Additional FastPrep history: Jul 19, 2024 |
ML Knowledge and Domain Questions
| Question | Frequency | All last seen dates |
|---|---|---|
| Depth on the team's areas: which equation fits a given situation, whether you know some lesser-known RL algorithms, and personalization models | 1 report | Feb 2026 (interview date) |
| What you've done with LLMs so far, and the LLM research topics you'd want to pursue | 1 report | Feb 7, 2026 (report date; interviews late 2025) |
| For a DoorDash business area matched to your background, how the methods from your projects would carry over, and which loss function to use | 1 report | Feb 3, 2024 (report date; interview Dec 2023) |
| Data drift and multi-task learning (the candidate wasn't sure these came up) | 1 report | Jan 24, 2024 (report date; interviews Oct–Dec 2023) |
Behavioral (BQ)
| Question | Frequency | All last seen dates |
|---|---|---|
| A challenging project that failed | 1 report | Sep 28, 2025 (report date) |
| A challenging project that succeeded | 1 report | Reports: Sep 28, 2025 (report date); Additional FastPrep history: Apr 1, 2026; Feb 19, 2025 |
| A previous failure | 1 report | Reports: Feb 3, 2024 (report date); Additional FastPrep history: Mar 22, 2026; Jul 19, 2024 |
| How do you handle conflict? | 1 report | Reports: Feb 3, 2024 (report date); Additional FastPrep history: Nov 5, 2026; Mar 22, 2026; Mar 24, 2025; Oct 9, 2024; Sep 3, 2024 |
| A very urgent deadline | 1 report | Feb 3, 2024 (report date) |
| Work you disliked and how you handled it | 1 report | Feb 3, 2024 (report date) |
9 detailed candidate timelines, loops, and screens
01: PhD MLE intern whose coding hour was a data-cleaning task with an AI assistant
Report date: Nov 7, 2026 (interviews Oct–early Nov 2026) · Role: Machine Learning Engineer Intern (PhD) · Team: New vertical consumer · Result: Pending; with a competing offer in hand, the candidate asked HR to hurry, and HR promised an answer within about two days
Online assessment
- The candidate took an online assessment first and was then invited straight to the virtual onsite. The assessment isn't described.
Virtual onsite, round 1: AI-assisted coding (1 hour)
- A simple dataset build rather than the workflow-engine task the candidate had expected. Five small tables about orders, IDs and stores had to become one training set on HackerRank: pick features from the existing columns, attach a label (as the candidate recalls, whether an order was placed), and handle duplicate rows and missing values along the way. Practice: Build a Training Dataset from Order Tables.
- Setup: an AI assistant (Codex, medium effort) inside the editor. The round also included a short discussion around the prompt and a read-out of what the results meant.
Break (15 minutes)
- A short break between the two rounds.
Virtual onsite, round 2: "domain knowledge" (1 hour)
- The candidate found the whole round behavioral: stories from their own work filled the hour, and research examples sometimes drew technical follow-ups.
02: Summer 2026 MLE intern: two interviews, then a move to AI Labs
Interview date: Feb 2026 (both interviews around Feb 10) · Role: Machine Learning Engineer Intern, Summer 2026 · Team: Interviewed for Ads Promo, placed on AI Labs · Result: Offer, around Mar 9, 2026
Application
- She applied within about an hour of the LinkedIn posting going up in late January and had an interview invite roughly an hour later. She says DoorDash had dropped the online assessment, so a résumé that passed the screen went straight to interviews.
- The invite came with a document explaining what each interview looks for. It also named the team she was being considered for, Ads Promo, and its focus areas: personalization, fine-tuning, reinforcement learning (RL), post-training and generative AI. It said she might be asked about her applied or research ML work and her fine-tuning experience.
Interview 1: coding (60 minutes)
- One medium-level, LeetCode-style problem in a DoorDash setting, about Dashers or delivery supply. She didn't name it. She says interviewers sometimes change the statement partway through to see how you adapt.
- The hour leaves about 40 minutes for the problem and about 5 for questions at the end. The editor was HackerRank-like. The interviewer nudged her toward the approach he had in mind, but she took a different route and he accepted it. She finished early, and they spent the spare time on DoorDash culture.
- She understood the bar as clear communication, speed and an optimal solution; in her view, passing all the tests is what counts.
Interview 2: engineering values and ML domain knowledge (60 minutes, no coding)
- The outline she was sent: about 5 minutes of introductions, about 40 minutes of ML discussion, a case study, and engineering-values (behavioral) questions inside the same hour. She didn't describe the case study or the behavioral prompts.
- The interviewer had a PhD in RL and kept pushing to find the edge of what she knew. He asked which equation would fit a given situation and whether she had heard of certain lesser-known RL algorithms. On one item she said she wasn't sure and gave her best guess; he told her that was fine because he wanted to find her limit.
- When personalization came up, she brought up what she called a "two-tier" transformer architecture, most likely the two-tower setup (our reading: one network for users, one for items). The interviewer said that's what DoorDash uses.
Team match (about 30 minutes)
- About two weeks later she heard the Ads Promo spot had been filled. Because her feedback was strong, she was matched with AI Labs (generative AI work) instead.
- The call with the AI Labs engineering manager was a deep dive into her projects and her Master's coursework. She saw it as a fit check, and the offer email followed soon after.
She had practiced with NeetCode 150 and DoorDash-tagged LeetCode questions; none of those came up. She also interviewed for the DoorDash software engineering internship in December 2025 but didn't describe it.
03: PhD MLE intern: two onsite rounds in January, a team match, then an offer
Interview date: Jan 13, 2026 (reported Feb 17, 2026) · Role: Machine Learning Engineer Intern (PhD) · Process: online application (Oct 2025), interviews arranged by a recruiter in December, team-match call Feb 5 · Result: Offer, Feb 12, 2026
Virtual onsite, round 1: coding
- A non-LeetCode task built on strings and dictionaries (hash maps). The exact task isn't given.
Virtual onsite, round 2: ML case analysis
- An ML case; its topic isn't given.
04: PhD MLE intern phone interviews: LLM questions, a search-recommendation design and a batching problem
Report date: Feb 7, 2026 (interviews late 2025, after Thanksgiving) · Role: Machine Learning Engineer Intern (PhD) · Stage: two phone interviews · Result: No offer; no contact from DoorDash afterwards
Round 1: team interview
- Background questions: mostly the candidate's LLM experience at the start, then résumé items and research interests in LLMs. The team described its work as LLM research.
- Design task: query-based restaurant recommendations for DoorDash, followed by a discussion of where an LLM could help. Practice: Design Personalized Restaurant Search.
Round 2: coding
- Minimum batches: group time points into batches of at most B points each, with every point within W of its batch's first time, and return the fewest batches. Approach: one greedy left-to-right pass, filling each batch before opening the next. Practice: Minimum Batches Within a Time Window.
- Format: runnable code, the candidate's own tests, and questions on possible improvements.
- Follow-up: Meeting Rooms II (LeetCode 253) as a rider-count problem; a sweep line, discussed but not coded. Practice: Minimum Riders for Delivery Intervals.
05: PhD intern phone interviews: a project round and a recursion-and-DP coding round
Report date: Sep 28, 2025 (interviews Jul–Sep 2025) · Role: Machine Learning Engineer Intern (PhD) · Stage: two phone interviews (order not given)
Projects and behavioral
- A detailed tour of past projects.
- Behavioral: a challenging project that failed and a challenging project that succeeded.
Coding
- An unnamed problem, solved with recursion and dynamic programming.
06: Fall 2025 ML co-op with a recommendation design question
Report date: Sep 1, 2025 · Role: Machine Learning Intern (Fall 2025 co-op; Master's degree required) · Location: San Francisco · Result: Offer (may have been declined)
Technical skills interview (1 hour)
- Basic ML knowledge questions; not listed.
- An open-ended design question: how would you build a recommendation system for food delivery? Related follow-up questions came after it. Practice: Model Restaurant Recommendations.
Break
- A 30-minute break between the two interviews.
Coding session (1 hour)
- Coding on trees and simulation. The specific problems aren't given.
07: Summer 2024 MLE intern: a coupon-credit problem and a domain hour built on the candidate's background
Interview date: Dec 2023 (reported Feb 3, 2024) · Role: Machine Learning Engineer Intern, Summer 2024 (Master's/PhD opening; PhD candidate) · Process: online application plus a LinkedIn message; no online assessment · Result: Did not pass
Hour 1: coding (60 minutes)
- Maximum coupon credit, rated easy by the candidate: each coupon has a value and one valid time range within a day (one example: 13:00–15:00, worth 10); return the largest total value of coupons valid at the same moment. Practice: Maximum Coupon Credit and Its Time Windows.
- Follow-ups: the time range or ranges that reach the maximum; time and space complexity.
- Format: the candidate's own test cases, with little attention to edge cases; one or two behavioral questions and Q&A in the spare minutes.
Hour 2: engineering values and domain knowledge (60 minutes, back to back with hour 1)
- Domain knowledge: two business areas matched to the candidate's background (not disclosed); for the one chosen, how methods from the candidate's projects would apply to DoorDash's use case and which loss function to use.
- Behavioral: a previous failure, plus one the candidate no longer recalls.
08: Master's MLE intern onsite: Longest Increasing Path and a case-style ML round
Report date: Feb 3, 2024 (interviews Oct–Dec 2023) · Role: Machine Learning Engineer Intern (Master's) · Stage: virtual onsite, two one-hour rounds back to back
Round 1: coding
- Longest Increasing Path in a Matrix (LeetCode 329); follow-up: a second algorithm for it. Practice: Longest Increasing Path in a Matrix.
Round 2: ML domain knowledge
- A case-style question set in a given business context: feature handling, and how to evaluate a new model's performance. The context isn't named.
Behavioral (both rounds)
09: PhD MLE intern phone screen: a restaurant-distance grid and two ML cases
Report date: Jan 24, 2024 (interviews Oct–Dec 2023) · Role: Machine Learning Engineer Intern (PhD) · Process: referral; phone screen · Result: Did not pass; the candidate blames the coding part
Coding
- Best starting cell, described only roughly: on a 0/1 grid where 1s are restaurants, find the cell with the smallest Manhattan distance to all of them. The candidate knew of a similar LeetCode problem. Practice: Best Starting Cell for Restaurant Distance.
Hiring manager: ML cases
- Two cases, recommendations and fraud detection, each asking for features, a model and evaluation metrics. Closest practice: Model Restaurant Recommendations and Model Order Fraud for Review.
- Possibly also data drift and multi-task learning; the candidate wasn't sure.
Patterns across the reports
-
One coding round plus one other conversation. All nine loops had a coding round. In five (01, 02, 06, 07, 08) coding took its own hour beside a second hour on ML, domain knowledge or behavioral questions; 03, 04 and 05 also had one coding round and one other round, with no lengths given. 01 had a 15-minute break between the hours and 06 a 30-minute one, while 07 and 08 ran the two back to back. 09 reports a coding problem and the hiring manager's ML cases without saying how the time was split.
-
Behavioral questions without a round of their own. Five loops had them (01, 02, 05, 07, 08), and none had a round billed as behavioral only. In 07 they filled spare minutes of the coding hour and shared the second hour with domain questions; in 08 both rounds had some; in 05 they came with the project walk-through; in 02 they shared the ML hour; and in 01, by the candidate's account, they filled the knowledge hour.
-
Recommendations from several directions. Four loops touched them (02, 04, 06, 09). 04 had to design query-based restaurant recommendations, then discuss where an LLM could help; 06 was asked outright to design one for food delivery; 09 had to pick a recommender's features, model and metrics; and in 02 the candidate brought up the model while discussing personalization.
-
Cases built around features and evaluation. In 08 the case asked how to handle features and how to check a new model; in 09 both cases, recommendations and fraud, asked for features, a model and evaluation metrics. 03's case and 02's were not described.
-
Problems about time, with the candidate's own tests. 04 packed time points into windowed batches, then counted riders for overlapping orders; 07 found the most coupon credit usable at one moment. Both candidates wrote their own test cases, and in 04 the code had to run.
-
Grid problems in late 2023. Two loops from Oct–Dec 2023 used a grid: the longest increasing path in a matrix (08) and the best starting cell for reaching every restaurant (09).
-
A team match before the offer. In 02 and 03 an offer followed a team-match call. In 02 the first team's spot was filled and the candidate was matched elsewhere about two weeks after the interviews; in 03 the call came about three weeks after the onsite, with the offer a week later.
-
The online assessment. 01 took one before its late-2026 onsite. 07 had none in December 2023, and 02 said DoorDash had dropped it by February 2026. The other loops don't mention one.
Keep practicing
- Coding: practice versions of reported problems: Minimum Batches Within a Time Window and Minimum Riders for Delivery Intervals (report 04), Maximum Coupon Credit and Its Time Windows (report 07), Longest Increasing Path in a Matrix (report 08), and Best Starting Cell for Restaurant Distance (report 09). The DoorDash question bank lists every DoorDash practice question in one place; it needs sign-in.
- AI coding: Build a Training Dataset from Order Tables is the practice version of the data task in report 01. More hands-on DoorDash projects are in the DoorDash project coding bank (sign-in needed).
- Design and ML cases: Design Personalized Restaurant Search (report 04), Model Restaurant Recommendations (reports 06 and 09), and Model Order Fraud for Review (report 09). More are in the DoorDash system design bank and the DoorDash ML modeling bank; both need sign-in.
- Behavioral: the DoorDash questions linked in reports 05, 07 and 08, and the rest in the DoorDash behavioral bank (sign-in needed).
- Everything DoorDash: the DoorDash interview page and DoorDash hiring insights.
- More firsthand reports: the Experience feed.
Sources: Anywhere useful interview info can be found.
Comments
0No comments yet — be the first to share what you know.