Palantir · OA → Phone screen → Onsite
In processPalantir FDSE New Grad OA Questions: 6 Experiences (Part 2 of 2)
Palantir FDSE new grad OAs and onsites from 2018 to 2023: three-part OAs with algorithm, SQL and REST tasks, a typosquatting test and decomposition.
At a glance
All six reports in this part are older than three years, so they sit in History below. For 2025–2026 reports, see Part 1.
- Process (2018–2023): OA on HackerRank → Karat technical screen (H08, H09) → recruiter call → one hour of decomposition and one hour of learning (H09).
- What they covered: two 2023 OAs each had one algorithm, one SQL and one REST API task (H06, H07); a 2020 OA had three cumulative typosquatting checks in 60 minutes (H10).
- Practise first: Network Formation, Count Fit Records Across Pages and Cumulative Typosquatting Detection.
Coverage: 6 Palantir Forward Deployed Software Engineer (FDSE / FDE) new grad candidates (reports 06–11), reported April 2018 to September 2023. They cover OAs, Karat, recruiter calls, learning and decomposition. One candidate moved from FDSE to SWE before the onsite (H08).
In this series
- Part 1: the newer reports, 2025–2026: screens, a map-and-accumulation learning round and an exoplanet decomposition.
- Part 2 (current): six older reports, 2018–2023: algorithm, SQL and REST OAs, an employee treasure hunt and the typosquatting OA.
Other Palantir cohorts: Internship · Experienced hires
Patterns across the reports
- One algorithm, one SQL, one REST task. Both 2023 OAs used this three-part shape (H06, H07), with different problems each time.
- Why Palantir again. It came up in the recruiter calls of H09 and H11.
Common questions
What is on the Palantir FDSE OA?
In 2023, three tasks: an algorithm, a SQL query and a paginated REST API task (H06, H07). In 2020 it was a 60-minute, three-step typosquatting checker (H10).
What happens in the Palantir FDSE onsite?
H09 had one hour of decomposition (an app to help employees meet) and one hour of learning (diagnosing package-fetching code with Futures), with 10–15 minutes of behavioral talk.
History (before Nov 2023)
These six reports are older than three years. They count toward this part but stay out of the summary tables.
H06: A photographed algorithm, SQL and REST assessment
Report date: September 7, 2023 · Role: Forward Deploy Engineer · Stage: OA · Result: Waitlisted
- Network Formation: form as many networks as possible from computers in their given order. Each network is a run of consecutive computers with at least
minCompsmembers and total speed at leastspeedThreshold; each computer joins at most one network, and some may be left out. 1–100,000 computers, speeds 1 to 1 billion, threshold up to 10¹⁴. Speeds[5, 7, 9, 12, 10, 13]with minimum 2 and threshold 15 give 2 (computers 2–4 with speed 28, and 5–6 with 23).[3, 3, 2, 5, 3, 2, 4]with minimum 2 and threshold 7 also gives 2. - Benchmarking Tool Report: from
devices(id, score), return one columndevice, ordered by ID, asDevice <id> has class: <class>. A is 80+, B 60–79, C 40–59, D 20–39, F below 20. Scores 20, 50, 50, 68, 95 map to D, C, C, B, A. - Count Fit Records Across Pages: call the medical-record REST endpoint from page 1, follow
total_pages, and count records whosevitals.bloodPressureDiastolelies in the inclusivelowerlimit–upperlimitrange. The original examples return 99 for 120–160 and 3 for 110–120 against the live API.
H07: Prefix sums, an employee join and population density
Report date: August 11, 2023 · Role: Forward Deployed Software Engineer · Stage: OA · Result: Waitlisted
- Extraordinary Substrings: the candidate said prefix sums solve it; the full statement and letter mapping were not shared.
- Employee Names With Access Records: return names of employees whose ID appears at least once in a second table. The candidate suggested an inner join with grouping; schemas not shared.
- Country Population Density: a REST task to compute a country's density by dividing two fields; response schema and rounding not shared.
- The candidate applied to FDSE; the invite was titled "Software Engineer Certified Assessment".
H08: FDSE stages, then a switch to SWE
Report date: January 24, 2023 · Role: FDSE, later SWE · Result: Not selected
- After an SWE resume rejection, a classmate referred the candidate to FDSE on November 15, 2022.
- They passed the OA and Karat and spoke with HR. No FDSE task was shared.
- Before the virtual onsite they were asked to switch to SWE and agreed; the candidate guessed headcount was the reason. The later project-depth questions belong to the SWE track. The post's January dates print 2022 after December 2022.
H09: An employee treasure hunt and a Futures discussion
Report dates: August 18 and September 12, 2022 · Role: FDSE, new grad · Result: Rejected after the virtual onsite
- Two posts by the same candidate, merged. The OA questions were forgotten.
- Karat: three time and space complexity questions, plus a coding problem with two or three parts of rising difficulty (domain not shared).
- Recruiter: why Palantir, recent projects and internship work, why FDSE rather than SWE, and other interviews or offers.
- Decomposition (one hour): build an app that helps employees meet. The candidate chose an employee treasure hunt where people combine clues. The interviewer liked it and asked why it would help, requirements and constraints, expected users and an MVP. The candidate sketched backend data structures, storage, UI and workflow, a database and how to scale.
- Learning (one hour): a machine fetches packages from several repositories. Given short code with multithreading and Future-like classes: what is wrong, how can it run faster in practice without changing Big O, and when would the original be faster? The concept was new to the candidate and they felt they answered poorly.
- Processes, threads and concurrency came up, and each round saved 10–15 minutes for behavioral talk.
H10: Typosquatting, each part keeping the earlier rules
Report date: November 5, 2020 · Role: Forward Deployed Engineer · Platform: HackerRank · Time: 60 minutes
- Applied in late October, then got "Typosquatting Version A" with three cumulative tasks: Cumulative Typosquatting Detection.
- Part 1: the same name under another domain suffix (
apple.comvsapple.org). - Part 2: also catch look-alike characters (
leetcode.comvsleetc0de.com). - Part 3: also catch one swap of adjacent characters (
palantir.comvsplaantir.com); non-adjacent swaps do not count, but the author countedpalanti.rcom, where the dot moves. - Each part had to keep passing the earlier cases. The candidate felt short on time.
H11: Contractor Fraud v4, then a recruiter call
Report date: April 10, 2018 · Role: FDE · Application: Referral · Status: Invited to a technical phone interview
- The OA was named "Detecting Contractor Fraud, v4"; no data model, rules or deliverable were shared.
- HR asked why Palantir and whether the candidate understood what an FDE does. They showed enthusiasm for customer work.
- A phone-interview invite followed; that round was not described.
Keep practicing
- Network Formation, Benchmarking Tool Report and Count Fit Records Across Pages: the H06 OA.
- Extraordinary Substrings, Employee Names With Access Records and Country Population Density: the H07 OA.
- Employee Treasure Hunt App: H09's decomposition.
- Cumulative Typosquatting Detection: H10.
- Palantir coding bank, SQL bank and system design bank: all Palantir questions by type (sign-in needed).
- Palantir interview guide and Palantir hiring insights: stages, practice and recent hiring signals.
- Experience feed: interview reports from other companies and roles.
In this series: Part 1 · Part 2 (current)
Sources: Anywhere useful interview info can be found.
Comments
0No comments yet — be the first to share what you know.