Interview kit

Interview kit from posted experiences

These are one person's posted experience, or a post passing along someone else's write-up. They are not this studio's interview, not a script, and formats change.

This page is separate from interview practice. That page holds the studio's own practice answers. This page only points at them. Reddit's own pages blocked a direct read from this studio, so the Reddit sections below follow the public post text archived for those same threads. The Medium article was fetched directly. Where a page was member-only, the locked part is skipped.

How to say it

I studied this fictional engagement. I have not been placed there.

A post about a Google forward-deployed interview

A July 2026 post describes a Google forward-deployed interview. It is written as a guide that mixes one loop with comments from other candidates. It is not one named person's first-person account. The post says there was a recruiter screen, then two technical conversations of about an hour, and no online assessment.

Vague enterprise scenario

The post says the first technical round was an enterprise agent design with a thin prompt: an automotive company wanted agents across several business functions, and many requirements were left open. It says clarifying the problem was part of the evaluation, and that people who drew architecture before they understood the business problem did worse. Later topics it lists include retrieval over internal documents, permission management, approval before a sensitive action, and a gradual rollout.

What to answer. Use Mill City Permits or Harbor Lane Clinic. Start by asking who the buyer is, the metric in their words, what is out of scope, and one written no. On Harbor Lane the written no is do not send, and do not read billing. On Mill City, pay stays out of the named corpus.

If the conversation turns to who may read a document, use week 3 and Harbor Lane. Billing and chart data stay out by a test, not by a policy sentence. The post lists permission management as one topic. It does not say that topic was the part someone failed.

If the conversation turns to rollout, use week 5 and Lumen Desk. Do not turn off the old path after a green check. The post only lists gradual rollout. It does not describe a repository migration.

Prefix search

The post says the second round was to return every dictionary word that matches a prefix, in alphabetical order, with an empty result when nothing matches, using a prefix tree, then to talk about scale, memory, and ranking by popularity instead of alphabet. That is a puzzle. This studio has no matching exercise, so there is no answer script for it. Do not treat the 45-minute page as that round.

A post relaying a Google phone screen

An Offer Engineering post says the author came across a Google forward-deployed phone screen. It names Google, a senior level, and Mountain View. It does not say the author sat the screen.

An agent that keeps acting

The post says the first part was a conversation about how an agent system should behave in production: scaling, safety, guardrails, monitoring, and what to do if an agent is stuck in a loop or keeps taking actions without progress.

What to answer. Use week 1. Name the step budget. Say the program, not the model, decides to stop.

A short conversion, then a repository migration

The post says the coding task was a small identifier conversion. The input was treated as valid. Once the format was clear, the code was short, and the coding was not the hard part. The follow-up was a large repository still using an old naming style: how to find rename candidates, avoid changing external calls, keep third-party calls, check the repository after the change, and handle tests. The post calls that discussion rollout safety, not a full design round.

What to answer. For the migration and the rollout, use week 5 and Lumen Desk. Do not turn off the old path after a green check. Do not invent a rename tool. The studio answer is the parallel run, not a script that rewrites a repository.

The posted coding task was an identifier conversion. This studio does not reconstruct it. If you want a practical coding answer that is not a puzzle, practice parse, join, and an idempotent write. Say that is what you studied. Do not say this phone screen asked for those three steps.

A Palantir forward-deployed post, public portion only

The August 2026 Medium article is member-only after a short opening. The public text says the author went through a Palantir forward-deployed process, did not advance, and believed the interview tested an ambiguous problem, the user behind it, making the problem concrete, clear reasoning, and a sensible implementation under time pressure. The next heading is about an initial phone screen. The body under that heading is not public, so that round is skipped. Nothing after the heading is used here.

What to answer. From that public description only, use Mill City or Harbor Lane. Ask who the buyer is, the metric in their words, what is out of scope, and one written no. There is no phone-screen script on this page, because the page does not show one.

A candidate's Palantir forward-deployed software engineer loop

A first-person post describes the author's loop for Palantir's forward-deployed software engineer role, on the government track. The author says that track does not sponsor and may require citizenship and a clearance. The author also says they were rejected, and that they believed the reason was mission alignment rather than the technical work.

Coding from a schema and from docs

The author says the coding screen was ordinary Python object design, not hard, with a strict cutoff around 30 minutes. A later round gave a database schema and documentation and asked for functions written on the spot. The hiring-manager technical portion felt like that same exercise. The author called it ordinary object-oriented work, not a trick.

What to answer. Practice the 45-minute page: parse, join, and an idempotent write. That is this studio's practical coding. It is not what the post says was asked. Say you studied the fictional orders. Do not claim the interview used them.

Decomposition from datasets

The author says one round was about 15 minutes of background and about 45 minutes of design. They were shown several linked datasets and diagrams and asked to design a system around them. Pseudocode was allowed. Diagrams were expected. The author describes it as logic and communication more than runnable code, including the shape of an API and the front-end flow.

What to answer. Use Mill City or Harbor Lane. Buyer, metric in their words, what is out of scope, and one written no. Draw that boundary before you draw a system.

Hiring manager

The author says the hiring manager asked about a summer internship and about why that company, why the government track, and why the role, and then described defense work the team sat inside. This studio does not script a mission answer. There is no case mapping for that part.

A candidate's Palantir forward-deployed software loop

A first-person post says the author had just finished a Palantir forward-deployed software loop with three rounds: coding, decomposition, and behavioral.

Two timed string and array problems

The author says coding was a third-party screen with two string or array problems in 60 minutes, a strict script, and no hints. That is a puzzle pair. This studio does not provide an answer for it. The 45-minute page is a different exercise, so it is not the mapping.

A vague product requirement

The author says the decomposition round gives a vague product requirement and asks you to turn it into a technical spec while someone watches: data models, API contracts, edge cases, and tradeoffs. The example in the post is a system that assigns analysts to investigations by expertise and availability. The author says this is closer to a planning document than to a classic systems-design prompt.

What to answer. Do not answer with that analyst desk. Use Mill City or Harbor Lane. Who is the buyer, what metric did they name, what is out of scope, and what is the written no.

Behavioral

The author says the behavioral round covered mission alignment, working with non-technical stakeholders, and ambiguity. The post does not describe a send button or billing in an index. For those two asks, use the answers already written on the interview practice page, including billing kept out of the index. Do not rewrite them. Say you studied the fictional engagement. You have not been placed there.

A write-up posted about Harvey

An Offer Engineering post shares an anonymized senior software-engineer write-up about Harvey. It does not name the candidate. It says the onsite had coding, an AI design round, and a project and values discussion, and that the candidate did not get an offer.

Citation highlighting

The post says coding was a string-processing problem about highlighting source citations inside generated text, and that this round was the manageable one. This studio does not reconstruct that problem. It is not the 45-minute page.

Answers over private documents

The post says the design prompt was a product where people ask questions across a large collection of private company documents and get answers grounded in those sources. The topics it names are document parsing and chunking, embeddings and retrieval, and reranking. The post then breaks off. It says the biggest challenge was access control.

What to answer. Use week 3 and Harbor Lane. Billing and chart data stay out by a test, not by a policy sentence. Filter who may read a page before rank. Do not name a model vendor in the answer.

Project and values

The post says the last round was past projects, ownership, technical decisions, collaboration, trade-offs, and lessons. The matching sentence is: I studied this fictional engagement. I have not been placed there.

A GenAI interview that became SQL and ETL

A first-person post does not name the company. The author says it was a large firm they had been trying to join, for a GenAI engineer role whose job text was also about GenAI. About the first 15 minutes were GenAI concepts, which felt comfortable. The interviewer then asked how good the author was with SQL. The next 30 minutes were ETL pipelines, data warehouses, data lakes, and data-engineering ideas. The author later learned the interviewer was a data engineer. The post does not say what happened afterward. Do not invent the company.

What to answer. Use Keel and Bin Parts and the reconciliation drill. The nightly file is not the live table. Show the gap SQL. Do not trust the nightly file.

What this page does not do

• It does not turn a post into a question the post did not describe.
• It does not give a prefix-tree solution or a string-puzzle solution.
• It does not fill in the Medium article past the public opening.
• It does not add a salary, a pass rate, or an offer the post did not state.