Want the $369K AI Role? Legora's Screen Filters 3 of 4 Applicants Before a Human Looks.

The Screening Funnel: How Legora Filters Applicants

Legora is hiring for 301 roles across four continents. The company's screen — a five-stage human loop — separates candidates who reach a case study with a hiring manager from those who hear nothing at all. That gap is where AI-native hiring lives, and it forces job seekers to adapt their applications to the filter.

Legora (founded in 2023 by Max Junestrand, Sigge Labor, and August Erséus) mirrors its product: a shared legal workspace where retrieval-augmented LLMs must cite every claim or the system discards the answer. That rigor shapes hiring. Glassdoor logs a 33-day average cycle across 82 reports, but the spread is wide: a Director of Sales Enablement candidate clears the full loop in three days; Customer Engagement roles average 60. The careers page cites two to four weeks; third-party guides put it closer to five.

The first gate is the application review. Legora's careers site promises updates within three to five business days. The company describes itself as in-office — 295 of 301 roles are on-site — so location alignment is checked early. Candidates who pass enter a structured interview sequence. Dataford's 77 reports show an overall difficulty of 4.7 out of 10, though the mix varies: 29 percent easy, 56 percent medium, 16 percent hard. The offer rate sits at roughly 29 percent (about one in four interviewees). Sentiment splits sharply: 40 percent positive, 47 percent negative.

Engineering roles follow a distinct track. The hiring-manager conversation and a case-study session drive the real evaluation, TechInterview's guide said. The case asks candidates to design retrieval over a firm's legal documents: chunking, indexing, ranking, and keeping answers tied to source passages. Interviewers grade on accuracy, citation, and hallucination prevention. A fabricated case or clause is an automatic fail signal.

Non-technical roles assess behavior differently but use the same cultural anchors: "mission before ego," "lean in," "fight for excellence," "grow together." Interviewers probe for adaptability, comfort with ambiguity, and craft seriousness — traits the company measures because the legal teams they serve "move fast, expect precision, and have no patience for vague answers."

Reapplication requires a cooling-off period. The careers FAQ asks candidates to wait at least a year before reapplying for the same role type. Relocation support exists depending on location, but the default expectation is on-site presence in metros including Stockholm HQ (55 roles), New York City (95), London (39), and a long tail from Chicago to Singapore.

Two Roles, Two Mandates

Legora's job board lists two openings as of October 2026:

The New York role carries the wider band and higher ceiling. The Stockholm role sits in the company's European hub, where Legora was founded and where much of the product engineering team remains.

The New York position maps closely to the AI Engineer specification Legora posted for its Stockholm office in September 2026. That posting describes a role inside "AI Enablement," a three-stream function that builds internal agents and tools, equips other teams to build their own, and owns the standards, access controls, and guardrails that let the whole company move fast safely. The team is small — "the mandate is large." That is the job, and the expectation is end-to-end ownership: architecture, production reliability, evaluation, monitoring, and whether anyone actually uses what ships. Candidates need strong production engineering fundamentals, deep experience with LLM-based systems (agents, retrieval, evaluation, tool use), and the ability to design for where those systems break. TypeScript is the daily language; Terraform and cloud providers are part of the work, not someone else's problem. Code agents are used daily, not as an experiment.

The same posting makes clear that Legora's internal users, about 1,000 people, most already fluent with current AI tools, will not adopt something worse than what they could build themselves in an afternoon. That constraint shapes the role: ship things people genuinely want, and the foundations arrive with them. The engineer must take a broad goal, work out what is actually required, break it into work, and argue for the priority order. The role demands three kinds of understanding at once (technical, domain, and business) with the ability to connect architectural and business conversations. The posting explicitly demands comfort with ambiguity: new team, fast growth, decisions that get revisited.

The Stockholm Senior Full-stack engineer listing is thinner on public detail, but the title suggests a product-focused full-stack contributor working on the customer-facing platform. Legora's customer base, more than 1,000 organizations across 50 markets including Cleary Gottlieb, Goodwin, Linklaters, White & Case, Dentons, and Barclays, runs on a product built with TypeScript across the stack. The Stockholm office anchors the company's European engineering presence, and the role demands the same ownership mindset: build production systems, not prototypes that need someone else to finish them.

Together the two openings reveal Legora's hiring priority: one engineer to harden the internal AI platform that powers the company's own velocity, and one to keep the customer-facing product moving at the same pace. Both roles demand production-grade engineering, TypeScript fluency, and the instinct to own outcomes end to end. The New York role adds the requirement to design knowledge architecture, permissions, evaluations, and risk-based guardrails at a scale where no specification exists yet. That is the sharper edge of Legora's filter — candidates who need a spec will not make it past the case stage.

The AI Layer: What Automated Screening Means for Legora

Roughly one in four employers now uses AI somewhere in their HR stack, and talent acquisition leads at 64 percent of those deployments, the Society for Human Resource Management said. Gartner found 76 percent of HR leaders believe their organization will fall behind without AI adoption within two years. The pressure to automate is real, and AI-native companies feel it acutely; they build the very models that power these screens.

The toolset is standardizing fast. Vendors bundle resume parsers, video-interview analyzers, candidate-success predictors, and scheduling optimizers into HR platforms that plug into existing applicant-tracking systems. A recruiter uploads a requisition; the software ranks the inbound pile, flags "high-fit" profiles, and surfaces a shortlist before a human opens a file. The mechanism is pattern matching against historical hiring data, and that is where bias enters.

Between 2014 and 2018, Amazon engineers built a resume scanner trained on a decade of the company's own applications, overwhelmingly male. The model learned to downgrade resumes containing phrases like "captain of the women's chess club" or degrees from all-women's colleges. Gender was never an input field; the system inferred it from proxy signals. Amazon scrapped the project, but the lesson persists: training data encodes past discrimination, and the model reproduces it at scale. Proxy variables operate quietly: zip code, undergraduate institution, commute distance, even video-interview cadence can correlate with protected traits. Foley & Lardner's 2026 briefing noted that "even small algorithmic biases arising from biased training data, flawed model design, or so-called 'proxy' variables... can produce selection ratios that may systematically disadvantage protected groups without any discriminatory intent."

Litigation is catching up. Derek Mobley, a Black disabled applicant over 40, sued Workday in 2023 after applying to more than 80 roles through its platform and receiving zero offers (often rejections within minutes). A Northern District of California court allowed his disparate-impact claim to proceed, writing that "drawing an artificial distinction between software decisionmakers and human decisionmakers would potentially gut anti-discrimination laws in the modern era." The same logic landed in Sirius v. XM Radio (2025), where Arshon Harper alleged 149 rejections by an AI hiring tool despite meeting listed qualifications. Large language models introduce newer hazards: a 2024 study found LLMs associate speakers of African American English with unemployment and lower qualifications, assign them less-prestigious jobs, and — in simulated sentencing tasks — impose death penalties more frequently. When a screening tool runs on an LLM backbone, those associations can bleed into candidate scores without any human reviewer noticing.

Regulators are moving in parallel. Executive Order 14110 (2023), the White House AI Bill of Rights, and laws in Illinois, New York City, Colorado, and California form a patchwork; no uniform standard exists. Employers bear the legal risk. Vendor assurances of "ADA compliance" or "bias-free" outputs do not shift liability. Foley's guidance is blunt: "Reliance on a vendor's assurance that the tool is 'ADA-compliant' does not relieve the employer of its independent legal obligations." Companies must inventory every AI touchpoint, demand validation studies and adverse-impact analyses, commission independent audits, retain model versions and decision logs, and maintain human oversight at consequential decision points.

The Human Layer: Interviews and Cultural Fit

The automated screen is only the gate. Once a candidate clears it, Legora runs a five-stage loop that takes roughly five weeks and ends with a conversation the CEO still conducts himself. First comes an application, then a People Team screen for fit and role alignment, followed by a hiring-manager interview weighted toward technical problem-solving, then a role-specific case study with a 60-minute presentation and team discussion, and finally a conversation with Max Junestrand or a department head.

Candidates encounter wide variance. Some describe a clear, structured process. Others report interviewers who appeared unprepared or unfamiliar with the candidate's background. Legal and Talent Partner applicants rate the interviews hardest; Data Analyst and Customer Manager candidates rate them easiest. For engineering roles, the hiring-manager and case stages drive the real testing: solid coding, system design grounded in LLM and retrieval pipelines, and behavioral questions about operating on a fast-moving team.

Junestrand's fingerprint is on the final stage. "I still interview everyone, so I ask quite brutal questions about, 'Why take a hard job? You could go work somewhere else,'" he told Business Insider in January 2026. His framing borrows from Kleiner Perkins partner John Doerr: "I try to create missionaries, not mercenaries. And I think we've successfully done that." The culture signal is explicit. At the company Christmas dinner, they served mulled wine — and KPIs. "We had the big sales dashboard in front of us at the wine thing, and everybody kept looking at it, because everybody wants momentum," Junestrand said. The company requires in-office work across Stockholm, London, and New York. Candidates should plan for on-site interviews and an in-person role.

The case study presentation carries disproportionate weight. Candidates present for roughly an hour, then defend design choices in front of the team. Interviewers weight clear reasoning heavily: stating assumptions, naming trade-offs, defending decisions. Engineering candidates should be ready to talk through a retrieval pipeline they have built or would build, how they would evaluate it, and how they would stop it from producing confident but wrong output. Legal-domain knowledge is not required; the founders entered without it. But genuine interest in the accuracy stakes of legal work helps a candidate stand out.

Headcount doubled in the last six months to 300, with plans to double again in the first half of 2026. That pace means interviewers are often new themselves, which explains some of the disorganization candidates report. Candidates who reach the final room with Junestrand should have a view on what they would build or improve, and a few real questions about the roadmap, the split between product teams and applied AI work, and how the company handles accuracy in a legal setting.

The Candidate's Playbook: Strategies to Get Past the Screen

Legora runs on the same automated plumbing as every other AI-native company. Its two open roles, the New York engineering role and the Stockholm full-stack position detailed above, will be parsed, scored, and ranked before a human ever opens the file. Nearly all recruiters use keyword filters in their ATS, and candidates whose resume title matches the target role are 10 times more likely to land an interview. Three out of four resumes are never seen by human eyes. If your formatting breaks the parser, your application dies silently with no rejection email and no explanation.

Start with the format. Single-column .docx or a clean text-based PDF. Standard fonts: Arial, Calibri, Georgia, Helvetica. No tables, no columns, no text boxes, no images, no headers or footers. Contact information sits in the body of the document. Section headings use the exact labels the parser expects: Summary, Experience, Skills, Education, Certifications. Clever titles like "Where I've Been" confuse the software and get your resume tossed. The 30-second test: select all, copy, paste into Notepad. If the output looks garbled, out of order, or missing sections, the ATS sees the same mess. Fix it before you submit.

Mirror the job description language exactly. Legora's AI Engineer posting asks for "production systems," "TypeScript," "Terraform," "cloud providers," and the LLM-based systems previously mentioned. Use those phrases verbatim where they honestly describe your experience. Do not paraphrase. "Cross-functional collaboration" and "teamwork" are not the same to a parser. Include both the acronym and the spelled-out version ("RAG (Retrieval-Augmented Generation)") because an AI looking for one may not recognize the other. Aim for 75% keyword overlap with the job description without copy-pasting it. Jobscan recommends that threshold.

Quantify everything. Recruiters and AI models reward the CAR method (Challenge, Action, Result). Weak: "Built backend services for legal AI platform." Strong: "Inherited 200ms p99 latency on document ingestion pipeline. Redesigned the queue architecture using Kafka and Rust. Cut p99 to 45ms while processing 50K docs/day." Numbers score higher than adjectives. "Managed high-volume annotation workload" scores lower than "Completed 300+ annotation tasks per week at 98% accuracy." Every bullet point needs all three components; the part almost every candidate skips is the measurable result.

Include a dedicated Skills section even if the same tools appear in your experience bullets. Parsers extract from a structured skills block more reliably. List technical tools, platforms, and methodologies relevant to the target role: Python, Rust, TypeScript, React, PostgreSQL, Kafka, AWS, Kubernetes, LangChain, vector databases, RAG pipelines, CI/CD. Soft skills carry more weight now; companies now train screening tools to detect interpersonal signals alongside technical ones. "Stakeholder communication" and "cross-functional leadership" in context carry real weight. Certifications need their full, standard names on first use: "AWS Certified Solutions Architect – Associate" before "AWS CSA-A."

Tailor the resume for each role. The New York position spans mid-level through staff; the Stockholm role is senior full-stack. A single resume cannot optimize for both. Create job-specific versions. Lead with domain expertise — legal tech, document processing, or LLM fine-tuning if you have it. Add any prior AI-related tasks even if informal: "Completed 200+ RLHF evaluation tasks on a platform" is a concrete, scoreable credential that generic CVs from other industries will not have.

Verify before you submit. The 10-point pre-submission check: file is .docx or text-based PDF; single-column layout with no tables, columns, or text boxes; contact info in the body; standard fonts; standard section headings; target job title appears in the professional summary near the top; keyword match above 75% without being a copy-paste; every bullet shows Challenge, Action, Result with numbers; certifications and acronyms spelled out on first use; plain-text preview test passed. Run your resume through the same kind of analysis the screening tools use. You'll find out in a minute what would otherwise take weeks of unanswered applications to learn.

The human layer still exists. A recruiter reviews the shortlisted resume in seconds. Keep it clear, specific, and scannable. Put your most relevant skills and measurable results near the top. A well-crafted note to a recruiter can help. Human connections matter. But you have to clear the algorithm first, and the case study waiting on the other side doesn't tolerate hallucinated answers. Junestrand still asks every finalist what draws them to a challenging role instead of a more comfortable alternative. The dashboard at the Christmas party wasn't theater. It's the operating rhythm. The screen selects for people who want to own that rhythm.

Working in AI? Zero G Talent tracks the openings: see every open Legora role, browse AI jobs, the companies hiring, and the people building the field.