company guides

Nous Research interview guide for open-source AI engineers

Nous Research is one of the few labs chasing frontier model quality in the open. It built its name on the Hermes models, open-weight releases that developers can download and run themselves, and on Psyche, a network for training large models across distributed, internet-connected hardware instead of a single owned cluster. That second bet is the unusual one: most labs assume a data center you control, and Nous is trying to show a competitive model can come out of coordinated machines spread across the internet. The company (careers page) describes its mission as democratizing access to the best intelligence, and it leans hard on being open where the large closed labs are not.

The money now says investors want that bet to pay off. Nous announced a $90M Series B in October 2026, with Nvidia, Samsung, M12, Union Square Ventures, Y Combinator, and Menlo Ventures named among the participants and Robot Ventures reported as the lead by some outlets (Pulse 2.0). Multiple outlets, citing the Wall Street Journal, pegged the valuation at roughly $1.5 billion; Nous did not confirm a number itself, and reports of its cumulative funding disagree, so treat the $1.5B as press-reported rather than official. The stated use of funds is to turn Hermes from a research artifact that developers download into a product businesses pay for, which is why several of the open roles point at product, deployment, and infrastructure rather than pure research.

Who Nous is hiring, and why the roles read the way they do

Nous lists its openings on its own careers page, and as of November 2026 the visible engineering roles are all remote: a Machine Learning Engineer focused on evaluations, a general Software Engineer application, Software Engineer roles for infrastructure, mobile, and product, a Forward Deployed Engineer, a Product Analytics Engineer, and a Data Engineer. The page notes a handful more roles that are not shown, alongside non-engineering openings like a product manager for the Hermes business and agent products and a partnerships lead. The shape tells you where the company is going. An evals-focused ML engineer and a data engineer exist because model quality is measured and defended, not asserted. The forward-deployed and product roles exist because Hermes is being sold to businesses now, and someone has to sit with those customers and make it work. One thing the postings leave open: they mark every engineering role remote but don’t state a country, a timezone window, or a visa-sponsorship policy, so if you would need sponsorship or sit outside a US-friendly timezone, confirm eligibility with the recruiter before you invest in the process.

For an engineer deciding whether to apply, the useful comparison is to other model-serving and applied-AI startups rather than to a big lab. The infrastructure and serving problems rhyme with what teams at Together AI, Fireworks AI, and Baseten work on, and the GPU-orchestration mindset overlaps with Modal. The difference at Nous is the open-source center of gravity and the Psyche angle: distributed training coordination is a systems problem most of those companies do not have to solve.

The interview process is not public yet

A November 2026 search of Glassdoor, Blind, and Reddit turned up no Nous Research interview account (the Glassdoor hits are for an unrelated firm, Nous Infosystems), and the company’s own postings describe the work and the application method, not the stages. So everything in this section is inferred from the roles and from how comparable early, mission-driven AI startups tend to run their loops. None of it is a reported sequence, and your recruiter’s agenda is the real source of truth once you are in the pipeline.

Because you apply by email rather than through an applicant-tracking funnel, the first filter is almost certainly a human reading your materials against a specific role, not a keyword screen. Any loop after that would be the usual Series B shape (a recruiter or founder screen, a hiring-manager technical round, deeper technical rounds, and a founder conversation), with the count, the take-home-versus-live format, and the timeline all unknown, so prep the substance and ask for the schedule in writing. Nous does set blunt expectations on its own page, telling candidates to expect “good wages, long months of complete focus, constant danger, with honor and glory in the event of success,” a Shackleton-style framing that signals intensity and a small team carrying real load.

What each track probably screens for

For the software and infrastructure roles, the problems are inference and training systems, so the questions that fit are systems-design under real constraints: where latency and memory go, how you reason about batching, caching, and failure. The multi-tenant, high-throughput problems in the system design interview guides are closer preparation than pure algorithm drilling, though a working command of data structures still matters for the coding portion. The Psyche side raises distributed-systems questions most candidates have not faced, so if you have genuine distributed-systems depth, say so early. The three prompts below are illustrative, built from the role shape rather than pulled from any candidate report, but they show the kind of question each track tends to turn on:

  • Serving and infrastructure: “You’re serving a 70B open-weight model behind an API with a p95 latency target and a fixed per-GPU memory budget. Walk me through batching and KV-cache management, and tell me what you trade off first when load spikes.”
  • Distributed training (Psyche): “Design the coordination for a training run spread across unreliable, geographically distributed nodes you don’t own. How do you handle stragglers, a node dropping mid-step, and keeping the run correct without a single cluster underneath it?”
  • Evaluations (ML engineer): “Build an eval for a new Hermes capability that resists contamination and gaming. How do you know the score reflects real ability and not memorized test data or a metric someone optimized against?”

The Machine Learning Engineer, Evals role lives on that third question: can you turn a fuzzy quality claim into a number a team will trust and defend when it gets challenged. For the Forward Deployed Engineer, the bar is split between engineering and customer judgment. You are expected to build real integrations and also to sit across from a business, translate a vague need into a working Hermes deployment, and own the outcome. Expect to walk through a time you shipped something directly for a customer and defend the trade-offs you made under their constraints. A structured STAR answer holds up better here than an improvised story.

Nous Research interview and company facts (as of November 2026) Detail How well-sourced
What it does Open-source AI lab behind the Hermes family of open-weight models and Psyche, its network for training large models across distributed, internet-connected hardware rather than one owned cluster. Now extending Hermes into a product for businesses Reported (company site and AI trade press)
Funding $90M Series B announced October 2026, with Nvidia, Samsung, M12, Y Combinator, and others participating and Robot Ventures reported as lead. Reported valuation near $1.5 billion (press, citing the Wall Street Journal); Nous did not state a valuation itself Reported (multiple outlets); valuation is press-attributed, not company-confirmed
Open roles Remote engineering roles visible in November 2026: Machine Learning Engineer (Evals), Software Engineer (general), Software Engineer for Infrastructure, Mobile, and Product, Forward Deployed Engineer, Product Analytics Engineer, Data Engineer. Additional roles noted but not listed; non-engineering roles include product manager and partnerships Reported (Nous Research careers page, read November 2026)
How to apply Email [email protected] with the role name in the subject line, attaching a resume or CV, a cover letter, and a portfolio of shipped work. No application form or formal ATS funnel is described; GitHub is not named, but public code is the natural portfolio for an open-source lab Reported (Nous Research careers page)
Interview loop No published loop and no candidate reports found. Plausible shape for a company this size: email-materials screen, recruiter or founder call, hiring-manager technical round, deeper technical rounds, and a founder or team conversation. Format, round count, and timeline unknown Inferred from roles and comparable AI startups; not reported. Confirm with the recruiter
Compensation No employer-stated pay bands on the Nous postings read in November 2026; a site like levels.fyi has thin Nous data. Equity carries most of the early-stage upside and risk No confirmed Nous bands; third-party estimates only. Ask the recruiter for your level’s band

Pay, equity, and what you can actually verify

The table says it plainly: no employer-stated bands, and the estimates floating around (job boards, a thin levels.fyi page) are outside guesses, not Nous numbers. So be skeptical of any single figure you see quoted, ask the recruiter for the band on your exact level before you invest in the loop, and weigh the offer as a package. At a Series B company, equity carries most of the potential upside and most of the risk, so run the whole thing through a total-comp calculator rather than comparing a base against a big-tech base. Once you have real numbers in hand, the usual salary-negotiation moves apply.

One practical note on applying: because the front door is a cold email, your materials do more work than they would in a funnel. An open-source lab reads shipped code as the strongest signal, so point to models you have fine-tuned, systems you have built, or contributions you have made, and make the “why Nous” line specific rather than generic. With no Nous reviews to study, the AI-native company interview guides and the AI-startup interview difficulty index are a better calibration for how a lab this size runs its loop, and the general AI-era interviewing advice about judgment over hype fits a team that measures itself against the closed frontier. When you do get a recruiter, send one direct email asking for the exact rounds, the format of each, and the pay band for your level. The answer will sharpen your prep more than anything you can infer from the outside.

newsletter

What's actually being asked right now

Interview patterns & comp trends, straight to your inbox.

No spam. Unsubscribe anytime.

newsletter

What's actually being asked right now

Interview patterns & comp trends, straight to your inbox.

No spam. Unsubscribe anytime.

1972 Soviet postage stamp commemorating the Mars 2 probe

worth a read

Mars For The Rest of Us — a weekly-or-more deep dive on the technical side of Mars exploration: rocket propulsion, microbiology, mission architecture, and everything in between. Written by Maciej Ceglowski.

Read it on Substack →
Scroll to Top