About this role
Electrical Engineer
verifinity.ai
Job descriptionApplication
About the role
Subject Matter Expert
Internal overview & hiring thesis We are hiring engineering Subject Matter Experts (SME) to turn real-world work into evaluations and task production for frontier AI labs. The purpose is to identify capabilities models cannot reliably perform today and build environments that help measure and develop those capabilities. The work should compound model capabilities and represent consequential engineering work performed at high volume. We think about opportunity at the level of real occupations and fields of work. An individual evaluation can focus on a constrained question or a substantial workflow, but it should fit into a broader picture of what an engineer in that field actually does. Across environments and connected tasks, we want coverage of meaningful professional work, including workflows that require multiple stages, tools, and sustained reasoning. The SME starts from curiosity about where models underperform in their specialization, investigates the failure, and turns that insight into tasks, worlds, and verifiers using Machina’s shared infrastructure and pre-training-scale data assets. Relevant environments may combine simulation and analytical tools, engineering applications, manufacturing equipment, or hardware such as FPGAs. We need people with deep fundamentals and the intellectual ability to investigate unfamiliar topics within their domain. They do not need to know every answer in advance. They do need to understand why an answer is right or wrong, recognize the limits of their expertise, and proactively close those gaps—including by finding the right specialist.
Role description & ownership As a SME, you will own evaluation design and production in your engineering domain. You will identify valuable capability gaps, transform real-world workflows into tasks and RL environments, reason through verification, investigate agent failures, and produce evaluations through Machina’s tools and infrastructure. Creating RL environments is part of the role. Machina’s infrastructure is intended to make verification development and quality checks straightforward enough that, in most cases, the lead does not need another engineer to build verification.
Each SME is a full-time domain expert who can scale specialist contractor teams for niche expertise.
Electrical & chip design The initial electrical focus is strong electrical and chip-design fundamentals, relevant ASIC or FPGA experience, RTL and design verification, EDA workflows, and understanding of synthesis, physical design, timing closure, and power, performance, and area tradeoffs. Real project experience carrying a design toward implementation is valuable evidence. Apply that expertise to engineering evaluations and RL environments; post-training experience is helpful rather than required.
Mechanical: product engineering The mechanical focus is product engineering, with substantial use of CAD and analytical solvers. The lead should bring strong engineering fundamentals and project experience that supports judging designs, interpreting analysis, and identifying where an agent’s engineering work goes wrong. Evaluations may involve applications such as SolidWorks and the physical or simulated assets supported by Machina’s infrastructure.
Materials: materials analysis The materials focus is materials analysis: characterization, failure and root-cause analysis, interpretation of experimental data, and measurement judgment. The lead should understand what evidence supports a conclusion about material behavior and be able to explain the reasoning clearly.
Success in the first 30 days
Within 30 days, the lead should be designing evaluations independently, creating tasks, and beginning to gain leverage by working with others.
The early expectation is that they can move from a meaningful model weakness to task design and production using Machina’s infrastructure, while recognizing where additional expertise is needed.
Responsibilities & representative work
Find capability gaps: investigate where frontier models underperform in the lead’s specialization and assess why improving those capabilities matters to real engineering work. Design evaluations and tasks: turn real-world workflows into constrained, technically meaningful questions and tasks that fairly test models. Connect individual evaluations to occupations and broader fields of work. Build worlds and RL environments: use Machina’s infrastructure, data assets, applications, simulation tools, and available physical equipment to make engineering work accessible to agents. Design verification: compare verification techniques and determine how correctness can be established for a particular task. Use Machina’s tools to build verification and run quality checks as part of production work. Diagnose model failures: inspect agent trajectories and explain the specific error, where the agent departed from a valid approach, and why the result represents a meaningful capability limitation. Develop task-production material: contribute task generation, synthetic recipes, taxonomies, and benchmark coverage for long-horizon engineering work, drawing on Machina’s pre-training-scale data assets.
Candidate profile What we are looking for
Deep engineering fundamentals and substantial intellectual curiosity. Can investigate unfamiliar questions, debug complex reasoning, and explain why an answer is correct or incorrect. Strong awareness of the limits of their knowledge. Proactively seeks evidence and knows when the right next step is to involve a specialist. Already uses AI models and agents aggressively and is enthusiastic about their potential, while holding a substantive view on their serious shortcomings. Owns the technical reasoning even when using agents. Can explain, inspect, and challenge the work agents produce rather than accepting it uncritically. High agency, care for the work, and inquisitiveness. Can make progress with minimal oversight and communicate the reasoning clearly enough for others to understand.
Preferred background & evidence Internships or a few years of engineering experience at elite organizations are preferred, rather than a mandatory filter. Publications and projects are valuable evidence, with particular weight on substantial project experience and the difficult problems the candidate has personally solved. Prior coding experience, evaluation design, and RL-environment-building experience are helpful, but are not prerequisites. Experience evaluating LLM capabilities in core engineering work and published work in engineering fundamentals are useful additional evidence. Post-training expertise is a bonus.
Interview & vetting
Before the assessment and interview, basic resume screener and other AI Assessment from Exordiom
- Digital assessment The candidate completes a trial assessment using Machina’s tools. Machina provides a written statement of the task requirements. The candidate may choose the type of evaluation and work within a field they know well. The central assignment is to design a constrained engineering question that fairly and reliably causes the state-of-the-art model at the time of the interview to fail, demonstrate a valid solution, and explain the failure precisely.
Tools, support & independence: Candidates can use Machina’s agents, tools, documentation, and other resources, but may not consult a specialist during the assessment. They can ask occasional questions. Machina’s documentation should be sufficient for a candidate to complete the work with very little oversight, and the team will review the logs of how they approached and completed the task.
- Live Screening Solution Review Zach or Dhaval will interview the candidate to discuss their solution. The write-up should make the task, valid solution, model attempts, and failure analysis understandable to Zach, Dhaval, and the team.
Explain the question, its constraints, the real engineering workflow it represents, and why the work matters to the world. Show a valid solution and the reasoning or evidence that establishes its correctness. Identify the specific failures from models and the exact point where the agent departed from a correct approach. Explain how the task could be verified and why the proposed verification approach is appropriate.
In the live discussion, the team works through the candidate’s solution and failure analysis in enough detail to understand the claims. The candidate must be able to decompose the subject and explain their reasoning to interviewers who are not specialists in that engineering niche.
Hiring decision criteria The quality of the failure matters. Defeating a model is meaningful only when the task is fair, technically valid, and represen