aishwaryanr
47ee6002b8
- Repoint 67 stale question-number references across the 4 prep plans to theme anchors (robust to renumbering) - Fix references to the removed role_based_prep.md, repointing to the hub
11 KiB
FDE Prep Plan
Concrete, day-by-day plans that sequence the rounds, questions, resources, and courses into a path. Two versions: a 3-week plan (about 2 to 3 hours per weekday) and a 1-week crunch plan.
Guiding principle. The FDE loop is roughly half technical and half judgment, and most candidates over-prepare coding and under-prepare the case, customer-simulation, and evaluation rounds. Weight your time toward the parts that decide the outcome: the ambiguous case round (highest weight, lowest pass rate), the customer simulation, and AI evaluation depth. Every session should end with something spoken aloud or written, not just read.
Before you start. Read the README and rounds once end to end so you know the shape of what you are preparing for. Pull the actual job description for your target company from Palantir Careers or the company's site, and note which rounds they run (see the company-specific shape section in rounds.md).
Timeline
3-week plan (about 2-3 hrs per weekday).
Week 1 |== Foundations + technical fluency ===========|
Week 2 |== Agents, evaluation, and system design ======|
Week 3 |== Judgment, communication, behavioral, mocks ==| --> ONSITE
3-week plan
Week 1: foundations and technical fluency
Day 1: Map the role and self-assess. Read README and rounds. Read the Exponent 2026 FDE guide and the Perspective AI FDE guide. Rate yourself 1 to 5 on each of the 10 question themes; the lowest two get extra time this week.
Day 2: Foundations. Work the Foundations topic. If shaky, read The Illustrated Transformer. Answer the GenAI foundations questions out loud, then refine.
Day 3: Prompting and context engineering. Do part of ChatGPT Prompt Engineering for Developers and read Anthropic on context engineering. Skim the Prompt Engineering Guide.
Day 4: RAG, part 1. Read the RAG topic and Agentic RAG 101. Answer the Retrieval and RAG questions aloud.
Day 5: RAG, part 2 (hands-on). Build a small RAG pipeline over a folder of documents (chunk, embed, retrieve, answer, cite), following a notebook from RAG_Techniques. Be able to defend your chunking choice with a retrieval metric.
Weekend: coding warm-up. Do 4 to 6 practical problems: parse messy CSV or JSON, a rate limiter, exponential backoff with jitter, a small CLI. Use NeetCode only to remove rust, not to grind algorithms. Narrate aloud as if an interviewer were watching.
Week 2: agents, evaluation, and system design
Day 6: Agents fundamentals. Start the Agentic AI Crash Course and read Building Effective Agents. Answer the agent fundamentals questions in the Agents and MCP theme aloud.
Day 7: MCP and tool design. Read the MCP introduction and skim the DeepLearning.AI MCP course. Answer the MCP questions in the Agents and MCP theme. Sketch how you would expose one of a customer's systems as an MCP server.
Day 8: Evaluation. Work through AI Evals for Everyone and read the Evidently AI LLM guide. Answer the Evaluation questions aloud. This is high-yield; give it real time.
Day 9: Deployment, cost, and reliability. Read the Production topic and skim LLM Numbers and the relevant Google SRE Book chapters. Answer the Deployment, cost, and reliability questions aloud.
Day 10: System design drills. Do 2 full system-design walkthroughs on a whiteboard, out loud, 45 minutes each, from the rounds.md design prompts (private VPC HIPAA RAG; 12 fragmented data sources; agent evaluation harness). Force yourself to start with an MVP walking skeleton and name trade-offs.
Weekend: responsible AI and security, plus review. Read the Safety and Security topic and Securing Agentic AI Systems. If targeting Anthropic, read Core Views on AI Safety and the Responsible Scaling Policy. Answer the Responsible AI and security questions. Re-answer any question you fumbled earlier in the week.
Week 3: judgment, communication, behavioral, and mocks
Day 11: The decomposition round, framework. Study the 5-step framework in rounds.md and read the Palantir FDSE interview guide. Do 1 full ambiguous case aloud from the Case and decomposition questions, narrating all 5 steps.
Day 12: Decomposition, reps. Do 2 more full cases aloud (another from the Case and decomposition questions and one design-case prompt from rounds.md). Time yourself to 45 minutes. Record yourself and check: did you clarify before solving, surface assumptions, sequence by risk and value, propose a thin MVP, and name failure modes?
Day 13: Customer simulation. Run the role-play scenarios in the Customer communication and business questions. Have a friend play the frustrated or non-technical customer if possible. Drill: diagnose before prescribing, acknowledge before pushing back, offer options with trade-offs, never over-promise.
Day 14: Behavioral stories. Write and rehearse 6 to 8 STAR stories covering the list in rounds.md: end-to-end ownership, difficult stakeholder, reversed decision, alignment without authority, tight deadline with incomplete info, a real failure, a cross-customer pattern, saying no and holding the line. 60 to 90 seconds each, "I" not "we." Prepare a specific, evidence-backed "why this company."
Day 15: Full mock loop. Simulate a compressed loop in one sitting: 1 coding problem, 1 system design, 1 decomposition case, 1 customer simulation, 3 behavioral questions. Get feedback or self-review against the "what good looks like" and "common mistakes" in rounds.md.
Weekend: targeted repair and rest. Re-drill only your weakest 2 rounds. Re-read the company-specific shape notes for your target. Skim your fumbled questions.md answers once more. Rest the day before the interview; do not cram.
1-week crunch plan
For when the loop is days away. About 3 to 4 hours per day, weighted hard toward judgment rounds and evaluation.
- Day 1: Orient and self-assess. Read README, rounds, and both FDE guides (Exponent, Perspective AI). Read all of questions.md once, marking weak answers.
- Day 2: AI depth. Speed-run the RAG, Agents, and Evaluation topic pages plus Building Effective Agents. Answer the Retrieval and RAG, Agents and MCP, and Evaluation questions aloud. Do not skip evaluation.
- Day 3: Coding and system design. 3 practical coding problems (parse messy data, rate limiter, backoff), narrated aloud. Then 2 system-design walkthroughs from rounds.md, starting each with an MVP skeleton and naming trade-offs.
- Day 4: Decomposition cases. The whole day on the signature round: study the 5-step framework, then do 3 full cases aloud from the Case and decomposition questions, recording and reviewing against "clarify before solving."
- Day 5: Customer simulation and reliability. Run all the role-plays in the Customer communication and business questions. Then read the Deployment, cost, and reliability and Responsible AI and security questions and answer the deployment, cost, and responsible-AI ones aloud.
- Day 6: Behavioral and company fit. Write and rehearse 6 STAR stories, "I" not "we," 60 to 90 seconds each. Prepare a specific "why this company" grounded in the company's real work. If targeting Anthropic, read Core Views on AI Safety.
- Day 7: One full mock, then rest. Compressed mock loop (coding, design, case, simulation, behavioral), self-reviewed against rounds.md. Light review of weak spots. Rest before the interview.
The night before, and the interview itself
- Re-read the "common mistakes" and "what good looks like" bullets for each round in rounds.md. They are the rubric.
- Internalize the one rule that fails the most candidates: clarify and scope before you solve, especially in the decomposition and design rounds. Silence and solution-jumping both lose points; narrate continuously.
- Use "I" not "we." Own your decisions and their trade-offs.
- Never over-promise in the customer simulation. Ownership language plus honest options beats confident guarantees.
- Have 2 or 3 sharp, specific questions ready for each interviewer about their customers and deployments; it signals genuine interest, which Palantir and Anthropic screen for.
Back to the FDE README.