RepostedPalantir logo

Palantir · Phone screen → Onsite

In process

Palantir FDSE OA and Interviews: 16 Experiences (Part 1 of 2)

Software engineer · UnknownReported Oct 31, 2026
Shared byjanebeingjane

Palantir FDSE and FDE reports from 2024 to 2026: a coding screen, then decomposition, learning and coding rounds, with a shared-interest app asked twice.

At a glance

  • Process: a recruiter or behavioral call, a coding screen (Karat for 60 minutes, or a shorter platform screen), then decomposition, learning and coding interviews of about an hour, often split 20/20 or 15/45, and a hiring manager (04, 05, 06, 07, 09).
  • Asked recently: decomposition or system design in 8 of 9 reports; coding in 7; a shared-interest app twice (06, 07).
  • Difficulty: candidates who solved the code still struggled with what the interviewer wanted (02) or with speed under pressure (03); coding ranged from LeetCode Medium (09) to a strict 30-minute object-oriented screen (05).
  • Practise first: Design a Shared-Interest Connection App, Find Unmet Users With Shared Interests and Merge a 2048 Row.

Coverage: Palantir Forward Deployed Software Engineer (FDSE / FDE) reports from September 2024 to October 2026: phone screens, Karat, coding, decomposition, learning, reengineering and hiring-manager rounds. Includes a Government new grad loop (05) and a candidate with 2.5 years of experience (09). No locations given.

In this series

  • Part 1 (current): reports 01–09, 2024–2026: combined screens, a Government loop, a shared-interest app and a 2048 row.
  • Part 2: reports 10–16, 2017–2022: Sydney, London, Munich and New York loops with War, recipes and ownership graphs.

The full series holds 16 reports; the seven older ones are in Part 2.

What came up

Palantir's FDSE loop is mostly about taking a loose product idea and making it concrete. Almost every report had a decomposition round, and coding was usually practical rather than puzzle-style.

Topic Reports Last seen
Decomposition or system design 8 of 9 Oct 2026 (report date)
Coding or implementation 7 of 9 Aug 2026
Behavioral and motivation 5 of 9 Aug 2026
Arrays, strings or interest lookup 3 of 9 Mar 2026 (report date)
Learning from documentation 2 of 9 Oct 2026 (report date)
Graphs (BFS, DFS) 2 of 9 Mar 2025 (report date)
Object-oriented coding 1 of 9 Dec 2025 (report date)
Debugging supplied code 1 of 9 Oct 2024

A candidate can appear in more than one row.

Questions we found

System Design and decomposition

Question Reports Last seen
Design an app that connects people by shared interests 2 Aug 2025 (report date)
Assign analysts to investigations by expertise and availability 1 Mar 2026 (report date)

Algorithms / Data Structures

Question Reports Last seen
Search for unmet users with shared interests 1 Mar 2025 (report date)
Merge a 2048 row 1 Jan 2025 (report date)

Behavioral (BQ)

Question Reports Last seen
Why Palantir? 2 Dec 2025 (report date)
Why FDSE? 1 Dec 2025 (report date)
Why computer science? 1 Dec 2025 (report date)
Why Government? 1 Dec 2025 (report date)
Your summer internship: what you liked most and least (practice) 1 Dec 2025 (report date)

9 detailed candidate timelines, loops, and screens

01: Learning first, decomposition a week later

Report date: October 21, 2026 · Role: FDSE · Status: Waiting

  1. The recruiter quickly set up the first onsite-stage interview: a learning round.
  2. More than a week passed with no news. After the candidate followed up, decomposition was booked for the next day.
  3. At posting they had done decomposition the Thursday before and had waited about another week. Neither task was described.

02: A finished solution, but a hard conversation

Interview period: August 2026 · Role: FDE · Result: Rejected

  1. The first call was expected to be HR but was with an engineer: a conversation about who they were, with no technical work.
  2. A later engineer interview: 20 minutes of decomposition and 20 minutes of coding. The coding was practical, not LeetCode-style, and they finished both. They had seen the decomposition prompt before but did not share it.
  3. What stuck with them was the conversation: they felt they never understood what the interviewer wanted, and clarifying questions did not settle it. A bug cost time before they finished the code.
  4. Gaps between interviews were three to four weeks. They blamed the rejection on those difficulties; no reason was given by the company.

03: Good marks for ambiguity, still a rejection

Report period: about spring 2026 · Role: Forward Deployed Software Engineer, final-year student · Result: Rejected

  1. The final loop started at 2 a.m. in their time zone and included demanding system design.
  2. Feedback praised their curiosity and how they worked through ambiguity. The gap, as they understood it, was faster independent algorithm analysis under interview conditions, which they called interview fluency.
  3. No technical question was named.

04: Two Karat problems and an analyst-assignment design

Report date: March 15, 2026 · Role: FDSE · Result: Offer

  1. Karat, 60 minutes: two string and array manipulation problems, both solved cleanly in their view. The operations were not shared.
  2. Decomposition: their approximate example was assigning analysts to investigations by expertise and availability. The talk moved from a loose product request to data models, API contracts, edge cases and trade-offs.
  3. Behavioral: mission alignment, working with non-technical stakeholders and handling ambiguity, without exact wording.

05: A Government new grad loop with a strict clock

Report date: December 13, 2025 · Role: FDSE, Government · Level: New grad · Result: Rejected

Recruiter and coding screen

  1. Motivation questions came back throughout: why computer science, why Palantir, why Government and why FDSE. After a pause, a recruiter follow-up restarted things about a week later.
  2. The technical screen was Python object-oriented coding with a strict 30-minute cutoff, about ten minutes of it talking. The class design was not shared. They advanced two days later.

Two virtual interviews

  1. First: about 15 minutes behavioral, 45 minutes decomposition. Connected datasets and diagrams set the scene; they were expected to discuss diagrams, APIs and the user experience, and pseudocode was allowed.
  2. Second: the same 15/45 split, but implementing functions from documentation and a database schema. They got stuck; the interviewer was patient and helpful. The functions were not shared.

Hiring manager

  1. After a next-day pass: about 30 minutes behavioral and 30 technical. The behavioral part covered their summer internship, what they did and liked most and least, and their understanding of FDSE. The technical part again meant implementing functions from documentation.
  2. The rejection came a few hours later. They wondered whether mission alignment was the issue; no reason was given.

06: A shared-interest app and a screen that felt good

Report date: August 27, 2025 · Role: FDSE · Result: Rejected

  1. Not a fresh grad, but without a solid one or two years of full-time SWE work. After the recruiter call came a 20-minute decomposition and 20-minute coding screen.
  2. Decomposition: an app that connects people through shared interests and activities. They sketched the user experience and schema, proposed a similarity metric, and covered APIs, messaging and invitations. They asked whether people might prefer group activities to one-on-one contact, and suggested scheduling group events around availability. The interviewer reacted well.
  3. Coding: not named. They used helper functions, handled edge cases and caught bugs; one late bug was treated as minor.
  4. An automated rejection came seven days later. They wondered about depth, experience, competition, or a question they asked about weekly hours; none was confirmed.

07: From a high-level app to a concrete search function

Report date: March 12, 2025 · Role: FDSE · Result: Rejected

  1. Behavioral call, 30 minutes, with a Deployment Strategist.
  2. Coding, one hour, about a week later: a BFS or DFS graph problem, not shared.
  3. Decomposition and implementation, about ten days later: an hour split roughly 20–25 minutes each. The prompt was an app that connects people through shared interests and activities. They opened with estimates, API and data flow, scale and security; the interviewer steered them back to what the app actually does.
  4. Then predefined classes and a search task, search(user_id, users). They asked about constraints and reached a working first solution with little help. The full return rule and class fields were not shared.
  5. Follow-ups: make meeting-history lookups faster than scanning a list (they proposed a hash map); improve interest lookup while keeping an array (they proposed sorting alphabetically and binary searching on the first character); and what other features the app could have (recommendations, category filters, interaction tracking).
  6. They did not advance.

08: A deliberately thin 2048 row prompt

Report date: January 28, 2025 · Role: Forward deployed SWE · Result: Not given

  1. A first coding interview on merging a 2048 row. The opening explanation was sparse, to invite questions.
  2. Examples: [2,0,2,0] -> [4,0,0,0], [2,2,4,4] -> [4,8,0,0], [2,4,2,4] -> [2,4,2,4].
  3. The work: move zeros to the end, double equal tiles that collide, then shift the new zeros. Input bounds and every edge rule were not given.

09: Five weeks and a 200-line debugging round

Interview period: September–October 2024 · Report date: November 1, 2024 · Role: FDSE, 2.5 years of experience · Result: Rejected before the hiring manager

Recruiter and phone screen

  1. A recruiter reached out in late September. A 30-minute call covered the resume and interest in Palantir.
  2. A HackerRank CodePair interview with an engineer: about 15 minutes on the resume and programming experience, then 35–40 minutes on a LeetCode Medium they solved with a hash map.

Three back-to-back technical interviews

  1. Coding: a time-series problem they rated medium to hard. They nearly finished in about 35 minutes; the engineer liked their explanation.
  2. Reengineering: over 200 lines of unfamiliar graph code. After a high-level walkthrough, they explained their thinking and found and fixed errors, though not all of them.
  3. Decomposition: a high-level real-world application, not shared. They felt it went very well.
  4. The rejection came three days later, about five weeks after the first recruiter message.

Patterns across the reports

  1. Decomposition paired with coding in one hour. Reports 02, 06 and 07 split one interview into about 20 minutes of each; 05 used 15 minutes behavioral and 45 decomposition.
  2. The shared-interest app repeats. Two candidates got it in decomposition (06, 07), and 07 then coded a search over it.
  3. Interviewers redirect toward the product. In 07 the interviewer pulled the candidate from scale and security back to what the app does; in 02 the candidate could not pin down what the interviewer wanted.
  4. Motivation matters. Why Palantir came up in 05 and 09; 04 and 05 both described mission alignment.

Common questions

What is the Palantir FDSE decomposition interview?

An open product prompt you turn into a design: user experience, data model, APIs and edge cases. Examples here are a shared-interest app (06, 07) and assigning analysts to investigations (04).

How long is the Palantir FDSE coding screen?

Karat ran 60 minutes with two problems (04). Other screens split 40 minutes into decomposition and coding (02, 06), or gave 30 minutes for object-oriented Python (05).

Does Palantir ask LeetCode questions for FDSE?

Sometimes: a LeetCode Medium (09) and a BFS or DFS problem (07). Others described practical coding, not LeetCode-style (02), or implementing functions from documentation (05).

What happens in the Palantir FDSE hiring manager round?

In the one report that reached it, about 30 minutes behavioral on the past internship and FDSE, and 30 minutes implementing functions from documentation (05).

Keep practicing

In this series

Sources: Anywhere useful interview info can be found.

2 total reactions. Your reactions: none.
0 comments

Comments

0

No comments yet — be the first to share what you know.