Parafin: Parafin is a fintech company that offers embedded financing, such as merchant cash advances, to small businesses through partner platforms.

## Pipeline

The software engineering loop is short: **technical screen → onsite**. The onsite combines a system design round, a coding round, and several behavioral / cross-functional conversations. Candidates turned down in an earlier loop have later been invited back straight into team-match conversations without repeating the loop.

## Technical screen

- A practical coding problem rooted in Parafin's own domain: merchant financing, sales-based repayments, fees, and similar money-flow bookkeeping. Expect structured input (businesses, daily figures, schedules) and a computed output, not a puzzle-style algorithm question.
- The same prompts recur, so problems already circulating among candidates are worth drilling.
- The interviewer is collaborative: when the logic has a gap, they nudge with hints and let the candidate patch it step by step. Talking through edge cases out loud pays off.

## Onsite

- **Location.** Bay Area candidates are invited to the San Francisco headquarters. Remote candidates may be offered an in-person option, but scheduling does not always work out; a virtual onsite is the fallback.
- **System design.** The prompt is built around Parafin's core business: embedded financing for small businesses through partner platforms. One concrete version is a loan application system. Requirements start deliberately vague, and clarifying questions narrow the scope sharply. Read up on what Parafin's products do (offer → application → underwriting → funding → repayment from sales) before the day. A standard framework (requirements → API → data model → high-level design → deep dives) works well here.
- **Coding.** Candidates choose a frontend or backend track. The backend version calls a real public API (for example a payments provider's API), fetches data, and processes it. The task itself is not hard; the risk is spending too long hunting for hidden traps and running out of time. The round closes with "what would you change before this code goes to production?" A good answer covers modularising the processing logic into components, error handling, retries, and tests.
- **Behavioral / cross-functional.** Separate conversations cover leadership and management, collaboration with product (a PM-led round), and resolving technical conflict within a team. Treat the PM round as a real evaluation, not a formality: show how you push back on and shape product requirements, not only how you execute them.

## Preparation

- Learn Parafin's product end to end, especially how merchant cash advances are offered, underwritten, funded and repaid as a percentage of sales.
- Practise a timed API-integration exercise: fetch paginated JSON from a public API, aggregate it, and handle errors, all inside 45 minutes.
- Prepare stories for leadership, PM disagreement, and technical conflict resolution.

Loop Structure & Round Formats

Timeline, Re-engagement & Visa

Parafin is a fintech company that offers embedded financing, such as merchant cash advances, to small businesses through partner platforms.

Amazon's bar leans on Leadership Principles and scenario thinking more than raw algorithm difficulty. Coding rounds skew toward OOD and under-specified prompts that reward clarification skill, and a GenAI-usage behavioral question appears in nearly every modern loop. Bar Raiser independence and writing-heavy work samples shape the senior-level signal.

Amazon

Apple interviews are unusually team-owned: the same company can run classic LeetCode screens, MLE case studies, hardware verification drills, project deep-dives, and system design loops with little advance disclosure. The strongest pattern is role fit: interviewers probe whether your prior work maps directly to the team's product surface, data domain, and engineering stack, while still expecting clean code, concise trade-off reasoning, and polished behavioral stories.

Apple

Uber's loop leans heavily on Uber Eats and rider/driver marketplace problems — Cart pricing, geo heatmaps, surge, and restaurant recommendation surface in nearly every onsite at L5+. Coding rounds are mostly tagged LeetCode patterns, but Hack2Hire OAs ship a small rotating set of original problems that rarely appear elsewhere. Coding bar is the gate; system design quality drives leveling between L5a / L5b / Senior / Staff.

Uber

Google's loop runs on hiring-committee arbitration, with a separate team-match phase deciding whether HC approval turns into an offer. Coding bars are remarkably consistent across teams; leveling rigor and behavioral depth drive most of the L3–L6 differentiation.

Google

ByteDance / TikTok runs a strict one-round-at-a-time funnel with sharp org-by-org variance in bar, language, and round mix. Coding rounds skew toward classic LeetCode-medium with one self-built test case; MLE and SRE loops front-load 30+ minutes of oral 八股 before any code is written. Interviewer professionalism varies — strong candidates drive the round structurally rather than waiting to be assessed.

Bytedance

Snowflake's loop is small and bank-driven: a recruiter screen, two back-to-back coding phone screens, then a virtual onsite that mixes coding, system design, and behavioral. The same handful of coding families (multi-source BFS on a grid, course-schedule variants, tree-node-deletion height problems, transactional KV stores) cycle through most candidates, so recognition speed matters more than novel problem solving.

Snowflake

Microsoft runs two parallel hiring tracks: the standardized Hiring Event (HE) loop covering Azure, Cloud, Copilot, and Office orgs, and Mustafa Suleyman's MAI org with its own recruiter pool and a distinctly OpenAI-flavored coding rotation. The HE bar is consistent — four 60-minute rounds back-to-back on a fixed weekday — with medium-difficulty algorithms and one system-design slot; MAI loops add project deep-dive depth and pull problems from an in-memory-DB / beam-search / streaming-decoder pool that does not overlap with classic LeetCode prep.

Microsoft

Karat conducts outsourced technical screens on behalf of hiring companies, so the round follows Karat's format rather than the hiring team's. Rounds are tightly time-boxed and multi-part, and reading speed on a long prompt or unfamiliar starter code matters about as much as coding speed.

Karat

Meta sits mid-transition between two interview loops in 2026. The legacy track still drives most candidates through CodeSignal-style OAs and tightly time-boxed coding rounds, while a new AI-Native loop layers in AI-assisted coding, AI-Enabled system design, and an AI-oriented behavioral round on top. Which track you land in depends on org and recruiter, not candidate preference.

Openai

Capital One's engineering loop is unusually predictable: a long CodeSignal OA gates everyone, then a single Power Day stitches four 45-60 minute rounds — coding, system design, case study, and behavioral — back to back. The technical bar is moderate but the case study and BQ rounds carry real weight, especially for MLE / Applied Research / Data Science where the business-context framing of a model decision is the differentiating signal.

Capital One

Oracle's loops split along two clear axes: OCI (Cloud Infrastructure) runs a fairly standard LC + system-design loop with heavy Indian interviewer rotation, while OHAI (Oracle Health, the post-Cerner business unit) leans into OOD/class-design problems built around healthcare scenarios. Coding bars are mostly easy-to-medium, but interviewer variance and post-loop down-leveling / ghosting are unusually common — the loop result often determines the eventual level, not just the yes/no.

Oracle

C3 AI runs a compact, hiring-manager-led loop: a short behavioral conversation with the manager comes first, followed by a two-round onsite of coding and system design where design interviewers push hard on deep dives. New-grad Forward Deployed Engineer candidates start with a HackerRank assessment that blends CS fundamentals, applied-AI judgment, and practical coding.

C3.ai

Block (Square, Cash App) screens with practical, multi-part CoderPad problems modeled on real product logic, where tests and clean refactoring between parts matter as much as the algorithm. Senior loops add a conversational round on AI-native engineering workflow, and money-movement teams probe correctness, legacy tech debt, and regulated-system judgment.

Square

Waymo loops mix nominally LeetCode-style coding with deliberately ambiguous, autonomy-flavored prompts and a signature 'data fluency' round that's hard to drill for. Coding bars are mid-hard but interviewers frequently swap in original problems framed around self-driving scenarios; ML and senior loops add a behavior-prediction-style ML system design round on top.

Waymo

Stripe favors long realistic prompts in a full IDE — integration, debugging, refactor — over LeetCode-style algorithms. System design is grounded in payments and regulatory reality, and behavioral signals weigh ownership and written communication as heavily as raw technical depth.

Stripe

LinkedIn's loop pivoted heavily in 2025 — Staff candidates now sit a packed two-round phone screen (coding + project retro) before a five-round onsite that includes an AI-Coding session in CoderPad's assistant pane. Behavioural depth is the dominant downlevel axis; the same coding bar that passes at Senior frequently caps a Staff loop with a downlevel offer.

LinkedIn

Airbnb's loop is famously consistent and famously slow. The coding pool is small and reused — a candidate who has drilled the canonical questions (split-stay, terrain water-drop, KV store, code review) walks in advantaged. Differentiation happens in the code-review round, the experience deep-dive, and the standalone core-values round, which is calibrated separately from the rest of the loop and serves as a real gate. Expect strong technical performance to land in a multi-week team-match queue rather than a fast offer.

Airbnb

Roblox interviews lean heavily on a small, stable repertoire of coding and system design prompts drawn from the company's own product surface — rate limiting, like-button counters, matchmaking queues, delayed payments. The bar is high but predictable: interviewers want bug-free implementations with crisp boundary handling and design discussions that ground every trade-off in real platform scale (millions of concurrent players, hot creator experiences, cross-region latency). HM and director rounds heavily probe past experience with comparable traffic and scope.

Roblox

Pinterest's loop leans hard on a recurring custom prompt library — escape-room state-tracking, trie-backed access control, ads/log aggregation, splitwise-style settlement — over canonical LeetCode. ML system design is split into two distinct rounds ("ML Practitioner" and "ML System Design") that effectively cover the same recommender / retrieval / ranking surface from two angles, with embedding-heavy product framing throughout. Fully-remote, front-loaded compensation, and long team-match windows define the experience as much as the technical bar.

Pinterest

Disney hires through individual organizations such as Streaming and Experiences, and each hiring manager shapes the loop, so formats vary widely even for the same title. Coding is mostly practical and moderate in difficulty; project depth, collaboration, and trade-off-heavy design discussions carry most of the weight.

Disney

Netflix's current engineering loop is high-signal and repeat-heavy: Ads candidates see the same topo / frequency-cap / ads data-modeling trio, Member-Commerce-Games candidates see homepage dedupe and duplicate-detection variants, and Infra / Data Platform candidates see cache, latency, and concurrency rounds. The technical prompts are usually solvable, but the grading is communication-heavy: candidates are expected to clarify vague requirements, write tests, explain trade-offs, and connect their past work tightly to the target org.

Netflix

Goldman Sachs interview loops lean heavily on traditional LeetCode-style coding — the firm publishes its own tagged problem list on LeetCode, and the OA / phone screens recycle from a stable bank that maps closely to it. Behavioral signal is concentrated in a Hirevue video screen with tight 30-second prep windows, and quant-flavored desks (GSAM, Quant Engineering) layer probability and statistics drills on top of the standard coding bar.

Goldman Sachs

Citadel and Citadel Securities run two distinct loops sharing one brand. Citsec (the market maker) hires for HFT software, quant research and quant dev; its phone screens lean on small, sharp coding prompts (merge-K, ring buffers, stock-DP ladders) and the loop is famously single-fail — any weak round and the next interview is cancelled within a day. Citadel (the hedge fund) and its sub-orgs (GQS, EQR, NXT) lean harder on probability brainteasers, deep resume drilling and a multi-week team-match stage where coding bar drops but cultural and PM-style fit dominates the offer decision.

Citadel

Salesforce's loop is bank-driven: a small set of coding prompts (LFU Cache, Flatten JSON, linked-list dedup, frontend curry + traffic light, OA HackerRank classics) and one signature system design (Coffee Ordering, with its SQL follow-ups) cover the bulk of recent rounds. Down-leveling at offer time is common — the HM screen and behavioural depth materially shape the final level, often more than coding correctness alone.

Salesforce

Bloomberg's loop is a coding-heavy gauntlet: three back-to-back technical rounds each pack two LeetCode-style problems with little hand-holding, plus an HR screen and an engineering manager round that anchors on a single deep project story. The coding bar emphasizes pristine communication — dry-running, edge cases, and complexity discussion are graded as heavily as the solution itself, and interviewers routinely push for second-best alternatives even after a working answer. New-grad and tech-screen verdicts often arrive within hours, while jump-shift onsites favor candidates who can explain past trade-offs without prompting.

Bloomberg

Nuro builds autonomous driving technology and hires engineers who are as comfortable with threads and locks as with algorithms: its coding interviews deliberately split into a classic algorithm track and an OS-and-concurrency systems track. System design leans toward the realities of an autonomous fleet — heavy batch compute and enormous volumes of vehicle data — and the process extends past the onsite into meetings with senior leadership.

NURO

Jump Trading interviews engineers as low-latency systems builders: after a Codility screen, technical rounds favour hand-built infrastructure — interpreters, concurrent data structures, allocators — over puzzle-style algorithms. Quant research interviews are open-ended, centred on data reasoning and an unusually deep interrogation of past research.

Jump Trading

Anthropic runs an unusually small, transparent technical question bank — the recruiter typically names which prompt family you'll see days before the round, so candidates can target preparation precisely instead of guessing. Bars are calibrated for problem-solving velocity and follow-up depth rather than novelty; offers are decided by hiring committee + reference check, and the culture round on AI-safety alignment is the single most common rejection point even for candidates who clear every technical stage.

Anthropic

Databricks interviews are unusually bank-driven: candidates who know the recurring prompt families often recognize the round within the first few minutes. The loop mixes algorithmic coding with low-level systems and product-grounded design, then puts real weight on behavioral, hiring-committee, and reference signals after the technical rounds.

Companies

Parafin Interview Question Bank

Members-only question

Members-only question

Members-only question

Members-only question