Everything in a Félix interview traces back to one fact: it moves other people’s money, and the WhatsApp chat is only the doorway. A worker in the US messages the Félix assistant on WhatsApp, and moments later money is on its way to a relative in Latin America, no separate app to install. Behind that surface sits a money-transmitter stack: funding the transfer, picking an FX rate, screening both parties, routing the payout, and reconciling every cent. Read the Series C structure before the headline: in September 2026 Félix closed $200 million, about $87 million of it equity led by Andreessen Horowitz, and roughly $113 million a credit facility from General Catalyst’s Customer Value Fund, not a second equity co-lead. Félix says it has processed over $8 billion in transfers and grew revenue more than 2.5x in the past year; the prior round was a $75 million Series B led by QED Investors in 2025.
The job description won’t say it, but Félix is a payments company first, a chat interface second. The assistant is the growth engine; the thing that can never break is the ledger. Interviews reward people who understand money movement, idempotency, and reconciliation more than people who can only ship a feature. If your background is consumer fintech, payments infrastructure, or risk systems, you’re the target; pure CRUD web work gets probed hard on the parts that matter.
The three engineering tracks, and how they diverge
The open roles cluster into three shapes. The payments-backend track is Python-first with Go, Kotlin, or Java on the services that need it, running as microservices on Google Cloud. The AI-agent track owns the WhatsApp conversation: intent handling, tool-calling, and the eval harness that keeps an assistant from doing something expensive, built on Google’s Agent Development Kit, Langfuse for tracing, and the OpenAI and Gemini APIs. The fraud-and-risk track is the one Félix has been hiring into hardest, with Software Engineer (Fraud) and Data Engineer III (Fraud Team) postings that ask for two or more years specifically on fraud or risk systems. Which of the three you land in decides your entire middle loop, so get the recruiter to name it on the first call. Public postings don’t show a separate frontend track; with a chat-first product, most product engineering is full-stack on the Python backend.
The stack, per a 2026 Félix Engineering Manager (AI) posting: Python with Flask or FastAPI, Redis, SQL, Pub/Sub, and the usual Google Cloud pieces (Firestore, Cloud Run, BigQuery, Cloud Composer), with Kubernetes and Terraform underneath. That same posting calls Félix technology-agnostic, so treat the list as a signal of taste rather than a checklist to memorize.
| Round (reasoned from real Félix postings and consumer-fintech peers, not a confirmed Félix loop) | Who runs it | What it likely screens for |
|---|---|---|
| Recruiter screen (30 min) | Recruiter | Track fit, fintech or risk background, work authorization and any sponsorship need, location and remote scope, comp expectations, possible Spanish-language expectations on cross-border teams |
| Practical coding (45-60 min) | Engineer | Working code in Python (or Go/Kotlin), data structures, a SQL question, clean handling of edge cases around money and retries |
| Deep technical dive (60 min) | Senior engineer | Backend: ledgers and idempotency. Agent: conversation design and evals. Fraud: feature pipelines and model tradeoffs |
| System design or case (60 min) | Senior/staff engineer | Design a transfer flow, a fraud-scoring path, or the WhatsApp message pipeline end to end, with failure modes front and center |
| Behavioral / mission fit (45 min) | Hiring manager | Ownership, working across the US and Mexico offices, and why serving immigrant remittances matters to you |
A Series C company hiring specialists tends to run four or five rounds. Reading the postings, domain depth seems to outweigh raw algorithm speed: a fraud or payments track record the technical round can confirm will carry you further than a perfect graph traversal. Treat that as a working theory, keep your fundamentals current with the coding patterns catalog, and go in ready to write real code.
The payments-backend round: idempotency and the ledger
On the money-movement side, the technical round lives on the questions that keep a remittance company solvent. The classic one: a user taps send, the request times out, they tap again, and you have to guarantee the money moves exactly once. Be ready on idempotency keys, deduplicating at the API boundary, and what you persist before ever calling an external payout rail. Ledger design is the other anchor. Know why a double-entry ledger with immutable, append-only entries beats mutating a balance column, how you reconcile your internal ledger against what partners actually settled, and what you do when the two disagree.
Messaging and consistency come up because the architecture is event-driven on Pub/Sub, which delivers at least once, so consumers must be idempotent and handle out-of-order and poison messages. A SQL question is fair game since SQL is in the stack: expect something like reconciling two transaction tables or finding duplicate or stuck transfers with a window function, so keep your SQL interview reflexes warm. FX is the fintech-specific wrinkle: how you quote a rate, how long you lock it, who eats the slippage when the market moves between quote and settlement. The through-line is that correctness and auditability beat cleverness, because every entry has to survive a regulator reading it back.
The AI-agent round: an assistant that moves money
The agent track is where the WhatsApp assistant lives, and the interesting questions are about control rather than clever prompting. How do you stop an LLM from confirming a transfer the user never authorized? Design the boundary between the model and the money: the agent proposes, but a deterministic backend validates and executes, with irreversible actions behind explicit confirmation. Tool-calling design, trustworthy structured output, and fallbacks for a wrong model are all fair.
Evaluation separates people who have shipped an agent in production from people who have only demoed one. That same Engineering Manager (AI) posting names Langfuse for tracing and lists F1 among its metrics, so be ready to say how you measure a conversational assistant that handles real transfers: offline eval sets, scoring intent classification with precision and recall, catching regressions before users see them, and watching conversion once it ships. Because the WhatsApp Business API is the front door, questions about webhook handling, delivery guarantees, rate limits, and templated versus free-form messages are reasonable.
The fraud and risk-data round
Fraud is where Félix has been hiring most visibly, and the round reflects it. Cross-border remittance is a magnet for money laundering, account takeover, and first-party fraud, so expect to discuss the screening path: KYC at onboarding, sanctions and OFAC checks on both parties, velocity rules, and where a machine-learning model sits versus hard rules. The modeling question that matters most is the precision-recall tradeoff. Block too aggressively and you freeze legitimate families sending rent money; block too little and you eat losses and regulatory heat. Be ready to reason about where you set the threshold and why, and to measure it with F1 or a cost-weighted metric rather than raw accuracy on an imbalanced dataset.
The data-engineering version of this round is about feeding those models: building feature pipelines in BigQuery and Cloud Composer, serving risk signals to backend microservices at low latency, and keeping training and serving features consistent. Expect questions on real-time versus batch scoring, backfills, and how you investigate a spike in a fraud signal.
Compensation: what’s grounded and what to ask
Félix is private and headquartered in Florida, which has no pay-transparency law, so a Miami-tagged posting may show no band at all while a role also open to New York or California carries one. A 2026 Engineering Manager (AI) posting for San Francisco and New York lists a base of $232,000 to $280,000 plus options and a bonus; a Technical Product Manager (Fraud) posting in Miami lists $123,000 to $160,000. The EM figure is a management-role band, used here only as an upper-bound anchor for senior individual contributors, not an IC range; the Miami number is a different role family in a different market, so don’t read the whole gap between the two as geography. To attach a real base number to a title, pull any DOL H-1B and LCA filings under Félix’s legal entity, which list exact wages by title and worksite. Whether Félix sponsors a given role isn’t stated in its public materials, so treat sponsorship as a recruiter-screen question. Part of engineering, including an Engineering Manager (Fraud) role, sits in Mexico and is priced in that market, so a remote-from-LatAm offer won’t track a US band. Whether the company hires remote from outside the Americas, and whether Spanish is required or just a plus, isn’t stated publicly either, so confirm both early.
Equity is the part you cannot value from outside: the company is private and disclosed no post-round valuation, saying only that the number tripled since the Series B (a figure near $1.4 billion in secondary coverage is unconfirmed, so don’t anchor a negotiation on it). So negotiate the terms that set the price instead of a headline number. Get the option count, the strike, the latest preferred price per share, the fully diluted share total, and where you sit in the liquidation stack; drop those into the total comp calculator to bound the range; and treat the grant and the base as two independent asks in any salary negotiation. For how a fintech loop and its pay compare across the sector, the Stripe, Mercury, and Plaid guides are useful reference points, keeping in mind that Félix is earlier-stage and consumer-facing.
The behavioral round and how to calibrate
The mission-fit conversation carries real weight here. Félix is built around immigrants sending money to family, and the people hiring care whether that lands for you, especially with teams spanning the US and Mexico. Expect to defend a hard technical call, describe a production incident you owned where money or trust was on the line, and show you can collaborate across time zones and languages. Keep the stories tight and specific with the STAR method. A design round tests whether you can hold the whole transfer flow in your head, so the broader system design material and the rate limiter pattern map onto a remittance API’s throttling and abuse problems. To see how this loop stacks against other early-stage shops, the AI-native company interview guides and the full company guides index are the place to calibrate. What earns the offer is specific: show, with real examples, that you know how money moves without going missing.
Practice the behavioral round:
