Imbue’s Two Roles Start With a 70% Keyword Test

Two Roles, One Bar

Imbue, formerly Generally Intelligent, which announced a $200 million Series B in 2023 and reached a $1 billion valuation that October, lists two open research positions: a Research Engineer and a Research Scientist, per Glassdoor. The company's public materials describe an interview process built as a "two-way street" that fosters "mutual understanding and alignment," asking candidates to be transparent so the team can understand them "as a whole person: from exploring your deeper goals, values, and motivations to emulating what it's like to collaborate with you." Imbue's technical hiring page acknowledges the design tension: "Designing a great interview process for engineers is tricky. How do we offer an accurate and comprehensive view of the role while respecting your time? How do we give you the opportunity to showcase your skills without resorting to contrived tests or introducing bias?"

The openings sit at the intersection of large-language-model training, agent architecture, and the infrastructure to run long-horizon tasks reliably. Imbue's open-source repositories, including mngr (a CLI for managing coding agents), Vet (code review for agent overconfidence), Darwinian Evolver (evolving prompts and code through iterative search), and Sculptor (a desktop workspace for parallel coding agents), signal the engineering surface area. Public statements emphasize demonstrated output, such as open-source contributions, reproducible repositories, or prior work at frontier labs, over credentials. Neither role requires a specific degree.

Market context: Figure AI, Boston Dynamics, and Zipline have recently posted roles drawing from overlapping talent pools, though each company's domain differs: AI agents, robotics, and autonomous delivery respectively. Zero G Talent's board data shows the table below summarizes their reported salary bands and medians.

Inside the Screening Funnel

The mechanics follow a structure familiar to high-volume technical searches. A 2023 walkthrough of standard hiring workflows shows the hiring team reviews every application, including resumes, cover letters, and structured forms, and asks a binary question: does this candidate meet the minimum qualifications to succeed? If no, the candidate exits immediately. If yes, the review moves to preferred qualifications: which ones does the candidate satisfy, and how many? Candidates who meet none are not rejected outright but are deprioritized; those with some or all are added to the interview rank list, ordered by the hiring team's eagerness to speak with them. Only after that ranking does outreach begin. A recruiter, hiring manager, or HR representative contacts top-ranked candidates to schedule interviews. The entire process, from application to offer, typically spans one to two months, sometimes longer, because of the number of steps.

Imbue has not published the exact sequence of technical screens, take-home assignments, or onsite loops for its two current openings. Academic literature on AI-enabled recruitment notes that automated resume screening can classify candidates at scale and reduce the pool recruiters must evaluate manually, yet warns that algorithms trained on historical hiring data risk replicating past biases — "bias in, bias out" — especially when datasets underrepresent certain groups. Imbue has not disclosed whether it uses algorithmic screening tools or relies solely on human review at the top of the funnel.

What is documented is the company's stated commitment to a process that feels collaborative rather than adversarial. The minimum-qualifications filter does heavy lifting; the preferred-qualifications sort then shapes the interview slate. Beyond that, candidates should expect a sequence designed to "emulate what it's like to collaborate," though the specific technical assessments and cultural-fit conversations remain undescribed in public sources.

Resume Signals That Clear the Filter

Applicant tracking systems score every resume before a human reads it. Research from Japply.io shows each application receives a match score based on keyword density, relevance, and recency of experience; candidates need a 70 percent-plus keyword match rate to consistently rank in the top tier. Keyword overstuffing backfires — modern parsers and recruiters both flag unnatural repetition — so the goal is precise alignment, not volume.

General guidance for AI research roles: vocabulary centers on large-model training and agent infrastructure. Resumes need distributed training primitives (FSDP, tensor parallelism, pipeline parallelism), checkpointing strategies, and experience with mixture-of-experts or retrieval-augmented generation at scale. Systems-track resumes need Kubernetes operators, GPU cluster observability (DCGM, Prometheus stacks), and failure recovery logic for multi-step agent trajectories. Both tracks signal depth with specific frameworks, such as PyTorch Distributed, JAX, or equivalent, not generic "deep learning experience."

Quantifiable outcomes separate deployed systems from lab demos. Hiring managers look for physical numbers: training throughput in tokens per second per GPU, model flops utilization, checkpoint recovery time, evaluation harness latency, and the number of distinct agent tasks a single checkpoint handles. A resume that says "improved training efficiency" loses to one that states "reduced 7B model training time 22 percent by rewriting the data loader to eliminate padding waste across 256 GPUs." The field generates measurable outcomes by nature; their absence suggests the system never reached production.

Formatting rules are non-negotiable. Submit as .docx or .pdf; creative formats, images, and tables can break ATS parsing entirely. Use standard section headings: Experience, Education, Skills. Place the target job title ("Research Engineer" or close variation) in the header or summary. Include both spelled-out terms and acronyms (Fully Sharded Data Parallel and FSDP; Reinforcement Learning from Human Feedback and RLHF). Front-load the top three to five keywords in the professional summary, which both ATS and hiring managers read first. Weave keywords naturally into achievement bullets (each bullet should contain at least one relevant keyword) rather than listing them in isolation. Mirror the exact phrasing from the job description: "activation checkpointing," "tensor parallelism," "agent trajectory evaluation" rather than paraphrasing.

Tailoring per application is mandatory. Each posting uses different phrasing; mirroring the exact keywords from each job description yields the best match score.

Technical Assessments: What the Simulations Test

Imbue has not published a public playbook for its technical evaluations, and no first-party board data or verified reporting details the exact challenges candidates face. The company's technical hiring page states it avoids contrived tests, but the specific assessments remain undisclosed. Candidates should prepare for evaluations that test fluency in the same stack the team uses daily, including distributed training, agent infrastructure, and the ability to debug real research bottlenecks, but the exact format is not public.

The Human Layer: Culture Under Pressure

Cultural fit interviews operate on different logic than technical screens. They test whether you will amplify or erode the team's operating culture when pressure mounts. A Top Interviews guide on cultural fit interviewing makes the stakes explicit: employers "don't just hire talent. They hire attitudes, values, and behaviors." A candidate whose skills are perfect on paper can still fail if their working style clashes with how the organization makes decisions, resolves conflict, or shares credit.

The guide identifies core dimensions interviewers probe. First, self-awareness: candidates must articulate what company culture means — "shared values, behaviors, and expectations that shape how employees interact and perform" — and then demonstrate how their own principles map to it. Generic agreement is insufficient. The guide's model answer frames alignment as daily embodiment: "I believe alignment means embodying values daily, not just agreeing with them verbally. By integrating integrity, collaboration, and continuous improvement into my work habits, I naturally support the organization's purpose."

Second, teamwork is treated as non-negotiable. The guide's scenario for handling unbalanced contributions ("I first focus on understanding the situation rather than making assumptions… initiate a respectful conversation to clarify responsibilities… involve leadership constructively") reveals the behavioral template: diagnose before acting, preserve cohesion, escalate only when fairness or productivity demands it.

Third, the ideal culture described by the guide's model candidate reads like a specification for a high-trust, high-autonomy environment: transparency, collaboration, continuous learning, psychological safety, diversity, and leadership that "communicates clearly and supports innovation while maintaining accountability." The guide argues that when employees feel "psychologically safe and aligned with company values, productivity naturally improves." This describes the conditions under which engineers can admit a failed experiment, challenge a senior architect's assumption, or propose a risky research direction without fear of performative punishment.

The guide's closing observation applies to both sides of the table: "Cultural fit interviews aren't about giving perfect answers. They're about showing who you truly are and how you align with a company's values. When you combine authenticity, self-awareness, and preparation, you stand out naturally."

[Note: The research provided does not contain Imbue-specific interview stages, leadership interview formats, or documented cultural values for Imbue. The above synthesizes general cultural-fit interview frameworks from the Top Interviews video.]

What This Story Does Not Cover

This article examines Imbue's hiring funnel for two open roles: how the screening stages work, which resume signals survive the first filter, what the technical assessments may test, and where cultural alignment gets evaluated. That scope is deliberate. Several adjacent topics, while relevant to understanding Imbue as an organization, fall outside this piece.

Compensation details for the two roles. The research does not disclose salary bands, equity ranges, or benefits packages for the specific positions Imbue currently lists. Zero G Talent's first-party board data shows salary bands for other AI and robotics companies, but Imbue's own listings are not in that dataset. Any compensation discussion here would be speculation.

The company's full funding history and cap table. Imbue raised $232M across three rounds from 22 investors, reaching a $1B valuation as of October 2023. The investor roster includes Drew Houston (Dropbox), Kyle Vogt (Cruise), Simon Last (Notion), Tim Hanson (Neuralink), Tom Brown (Anthropic), Jensen Huang (NVIDIA), Eric Schmidt (Google), and Michael Nielsen. Celeste Kidd, UC Berkeley psychology professor, also appears on the about page. This article does not analyze investor influence on hiring priorities, board composition, or how the $200M Series B (announced with the rebrand from Generally Intelligent) allocates to headcount growth.

Product roadmap and technical architecture. ** Imbue's public tools, including Bouncer (X feed filter), mngr (parallel coding agent CLI), Sculptor (desktop workspace for coding agents), plus open-source repositories like Vet, Darwinian Evolver, Keystone, Offload, Latchkey, and Jupyter Ascending, represent the engineering surface area candidates might touch. GitHub stars range from 395 (Bouncer) to 549 (self_supervised). The Microsoft Power BI roadmap items that appear in the research corpus are unrelated to Imbue and are not addressed. This piece does not evaluate product strategy, open-source maintenance burden, or how the agent ecosystem vision translates to day-to-day engineering tasks.

Advisor and leadership backgrounds beyond hiring relevance. The about page lists prominent advisors and supporters. Their individual research agendas, including Kidd's work on attention, curiosity, and cognitive development at UC Berkeley; Brown's GPT-3 authorship; and Vogt's Cruise trajectory, inform Imbue's cultural DNA but are not analyzed here. The article does not assess how advisor networks shape referral pipelines or technical direction.

Policy and advocacy work. Imbue states its policy team builds legal frameworks to protect individual liberty against concentrated AI power. That mission — "technology loyal to the user" — frames the cultural filter candidates encounter, but the specifics of policy engagements, regulatory filings, or lobbying activity are out of scope.

Physical workspace and team rituals. The San Francisco space offered free to "small teams building honest or punk software" signals cultural values. Whether the two open roles are office-mandated, hybrid, or remote; what onboarding looks like; how the team runs planning, retrospectives, or design reviews: none of that is covered.

Competitor hiring benchmarks. AMI Labs, xAI, and Letta are named as top competitors. Their hiring volumes, interview formats, or offer rates are not compared. The Zero G Talent board data for Zipline, Boston Dynamics, and Figure AI provides market context for robotics-adjacent roles but does not map to Imbue's AI-agent focus.

Historical hiring patterns. ** Whether Imbue typically runs two-role cohorts, how past classes performed, retention rates, or internal mobility paths: no data supports those questions.

Candidate experience narratives. No anonymized candidate accounts, recruiter interviews, or hiring-manager commentary were provided. The screening funnel described in earlier sections is reconstructed from public signals and structural inference, not first-person testimony.

Future funding or IPO timelines. The $1B valuation (October 2023) is the latest public marker. Subsequent capital events, revenue milestones, or liquidity plans are not addressed.

This article stops at the screening threshold: what gets a candidate through the door. The filter that matters is the one Imbue built — two roles, a multi-stage screen that passes engineers who have already internalized the constraints of shipping agents that act, not just chat. What happens after, including offer negotiation, onboarding, performance review, and career trajectory, belongs to a different story.

Working in robotics? Zero G Talent tracks the openings: see every open Zipline role, browse robotics jobs, openings at Boston Dynamics and Figure AI, and the people building the field.