Careers at Halluminate: Teams, Pay and How to Get Hired

Who Halluminate Hires

Halluminate pays a median $225K for its AI/ML roles. Founded in 2024 by Wyatt Marshall and Jerry Wu, the firm employs fewer than ten people but drives mid‑eight‑figure annual recurring revenue. Its hiring process favors applicants who show deep technical skill and fit with a fully remote, asynchronous, results‑oriented culture.

The firm creates reinforcement‑learning environments and benchmarks that pressure frontier models on knowledge‑work tasks, beginning with finance. To make those tools work, it staffs platform engineering, research and post‑training, forward‑deployed engineering, finance research, data infrastructure, and the people‑and‑recruiting systems that turn a sub‑10‑person team into an eight‑figure‑run‑rate operation. If your résumé reads like a generic AI checklist rather than a record of shipping domain‑specific tooling, you start behind the curve.

Halluminate advertised nine openings across operations, engineering, recruiting/HR, and sales in its latest board data.

Pay Structure

Halluminate splits compensation into two clear tracks.

Salaried track – The board‑observed band for salaried roles runs $150K–$275K. This covers core engineering and research hires based in San Francisco.

Expert track – Former investment‑banking, private‑equity, and consulting professionals work as 1099 contractors, earning hourly or milestone‑based pay for training and evaluating AI models on real financial work.

To avoid repeating numbers, the table below lists each advertised role with its salary band and location.

Verification drives the pay philosophy. Halluminate’s own tests show that shortcuts rewarded by models lead to sandboxes that agents escape, and real‑world validation yields a 0.6% success rate versus 23.4% (Home | Halluminate found) for mixed‑model approaches. The firm hires people who understand that gap and can build environments where it never appears.

The expert track pays top performers $225–$250 per hour; some contributors earn over $200K while still in school. A minimum of 20 hours per week for at least six consecutive weeks is required, with the ability to scale to 60 hours. Compensation arrives either as milestone bonuses of $1,500–$4,000 per project or as straight hourly rates that rise with performance. A salaried engineer at the top of the board-observed band makes $275K (FIRST-PARTY BOARD DATA reported). An expert billing 40 hours weekly at $250/hour pulls in $520K — nearly double. Because the expert track is 1099 only, it offers no visa sponsorship or W‑2 option, excluding H‑1B holders and those seeking traditional benefits.

Equity remains opaque. An Ashby listing from late September showed a Level 3 role (five‑to‑eight years experience) offering $250K–$320K in salary plus 0.2%–0.35% equity, above the board‑observed bands (Halluminate Jobs found). Glassdoor’s anonymous sample of two salaries echoes the board figures but lacks detail on tracks or seniority.

How the Hiring Process Works

Applications receive a review within three business days. Candidates first pick a role group — Private Equity, Investment Banking, Consulting, Accounting, FP&A, or Other, which sets the hourly‑rate baseline: PE and IB roles start at $150–$250/hour; the other groups begin at $100–$200/hour. This choice is not administrative; it anchors the pay scale from which performance‑based increases flow.

Selected entrants enter a paid two‑week training and onboarding period. During this immersion they complete required modules and deliver two successful problems for a milestone worth $1,500–$4,000, depending on the project. Work stays fully remote and asynchronous; the contract demands at least 20 hours per week for six straight weeks, though availability can push to 60 hours weekly. No fixed daily hours exist — schedules flex, including weekends by design.

Pay blends milestone and hourly components, rising quickly with demonstrated performance. Contractors can reach $225–$250/hour, and most strong contributors see an increase within the first month. Rate bumps tie to measurable outputs: number of problems delivered, total hours committed, responsiveness, and quality of communication. The highest rates go to senior writers who have delivered consistently over time.

Early interviews weight the ability to produce financial deliverables — Excel models, PowerPoint presentations, financial analyses, and to judge AI‑generated outputs against those benchmarks. The hiring bar represents the minimum standard across technical skills, problem‑solving, collaboration, communication, and role‑specific expectations. For finance‑domain expert roles, the core requirement is finance expertise, not engineering or coding. Candidates lacking a finance or accounting background face a steep climb, because the work involves building complex financial deliverables and then evaluating AI outputs against them.

The process values both deep technical ability and a fit with the company’s remote, asynchronous, results‑driven ethos. As 1099 contractors, participants must accept that classification; eligibility for CPT or OPT depends on individual school and visa terms. Halluminate cannot offer W‑2 sponsorship or H‑1B transfers.

Candidates who miss the hour commitment, cannot flag conflicts with current employers, or work from restricted countries do not advance. Those who skip the role‑group selection or cannot prove the required finance domain expertise also fail to receive an offer. Applications roll continuously; applicants may note availability around internships or other commitments, but the six‑week minimum and the 20‑hour‑per‑week floor remain firm.

From the three‑day review through paid training to performance‑based rate advancement, the pipeline seeks people who can train AI models on real financial tasks while thriving in Halluminate’s culture of flexibility, accountability, and results. For applicants, the path is clear: choose the role group that matches your background, prepare to showcase finance‑domain expertise, and commit to delivering the financial artifacts that become the benchmark data Halluminate uses to evaluate frontier models.

Where the Work Happens

Halluminate operates from a single headquarters in San Francisco, California, where its nine‑person team builds the reinforcement‑learning environments that train frontier AI models on finance‑specific work. The office blends a lab for environment design, a compute cluster for running benchmarks, and meeting spaces where domain experts from investment banking, private equity, and consulting shape the financial deliverables used to evaluate AI outputs.

In the lab, engineers construct computer‑use and tool‑use gyms that simulate tasks such as redlining a statement of work, rebuilding an Excel forecast, or drafting a PowerPoint pitch. The company’s own site notes that these gyms are built from anonymized private‑equity transaction data, emails, and meeting notes that mirror real deal rooms. The benchmarks they produce — like the World Finance Diligence Bench with 88 multi‑step problems and the DealTrace benchmark that traces every claim to its source, are assembled and tested here before being sent to partner model labs.

Adjacent to the lab sits a GPU‑based compute cluster that runs the reinforcement‑learning environments at scale. The cluster executes the hundreds‑of‑step trajectories required by the benchmarks, letting the team measure how often an agent omits a required change, relies on an outdated analytical method, or falls back on superseded information. As CEO Jerry Wu told Fortune, the complexity of these environments must roughly double every six to eight months to keep pushing frontier models — a pressure he calls the “Moore’s law of environments.” The compute cluster is upgraded accordingly to handle the longer trajectories and larger file sets that result from this doubling cadence.

Meeting rooms serve as the point of contact for Halluminate’s expert network. Professionals with IB, PE, and consulting backgrounds visit to create the complex financial deliverables — Excel models, PowerPoint presentations, and financial analyses, that become the ground truth for AI evaluation. After experts produce these artifacts, the AI‑generated outputs are compared against them in the lab, and any discrepancies flow back into the environment design loop. This tight coupling lets the company compound its finance expertise, expert network, and domain‑specific verification methods, as Wu explained in the same Fortune interview.

The office also stores the anonymized deal data that underpins the benchmarks. Because the data originates from real private transactions, Halluminate guards it with the same confidentiality standards used by the financial firms that supplied it. This storage lets researchers pull specific files, emails, or meeting notes on demand when they construct a new benchmark or modify an existing one, ensuring each training environment stays grounded in authentic deal workflows.

Overall, the San Francisco site functions as a facility where environment design, high‑performance computing, expert collaboration, and data stewardship intersect. Each physical element enables a specific part of the company’s mission: the lab turns expert knowledge into simulatable tasks, the compute cluster runs those tasks at the scale needed for frontier‑model training, the meeting rooms keep the expert network engaged, and the secure data store guarantees that the simulations reflect real‑world financial work. Candidates who picture themselves working in this integrated setting can better judge whether Halluminate’s hands‑on, environment‑focused approach matches their own strengths and career goals.

Who Thrives at Halluminate

Wu said Halluminate is profitable with only nine employees. Wu said four of the five leading closed‑source US AI labs pay Halluminate for its services. Consequently, the firm needs people who can own large chunks of work and move quickly from idea to prototype, while staying focused on the few customers that matter most.

Wu describes the systems Halluminate builds as infrastructure for “verticalized data research labs” that compound financial expertise and professional judgment into a training signal. Successful engineers treat each environment as a precision instrument. Subject‑matter‑expert review is baked into the verifier, and they prioritize alignment of task and reward with what a real expert would reward.

Halluminate’s benchmark tasks come from anonymized private‑equity transactions reviewed by active deal professionals, and the expert rubric becomes the verifier built into the environment. Thriving engineers show a habit of tracing every step of a long, messy workflow: document review and contract analysis, plus evolving term sheets and final deliverables. They notice where an agent loses track of intent, applies outdated data, or picks the wrong analytical method.

Because the team is small, each person wears multiple hats. A Member of Technical Staff might design the simulation core and write the expert‑annotated dataset. They might also help a customer integrate the platform, all in the same week. This demands comfort with ambiguity, a willingness to learn adjacent domains (finance, reinforcement learning, software engineering), and the ability to deliver complete features without waiting for hand‑offs.

The company’s profitability at nine employees reflects a concentrated, high‑value customer base that has no obvious substitute. Frontier labs have invested heavily in coding reinforcement‑learning environments, the highest‑demand category in post‑training infrastructure, but financial work has not received the same attention. Halluminate looks for people who can spot and fill that niche. It also seeks those who understand why financial due‑diligence is a hard verification problem, and who can build environments that teach models to carry instructions through to the end without drifting.

Alignment and safety rank as core concerns. Recent alignment incidents trace back to poorly built sandboxes and weak process verification, so Halluminate treats every environment as a safety‑critical system. Engineers who thrive here think about edge cases, build verifiers that catch subtle drift, and treat failure patterns as raw material for the next generation of environments.

Wu says the near‑term priority is to go deeper with existing frontier‑lab customers rather than chase enterprise buyers. This rewards a mindset of depth. Engineers who enjoy refining a small set of high‑impact projects fit this mindset. They also can iterate on complexity generation after generation and anticipate the next jump in environment scale: multi‑agent team simulations and simulated companies, potentially growing into full‑scale economies, as model intelligence doubles every six to eight months.

They also appreciate the investor’s perspective: Oak HC/FT, a fintech‑specialist backer, sees the work as financial‑services infrastructure, not pure AI capability. Thus, a background or interest in financial workflows adds concrete value. Experience in deal‑making or regulatory compliance does the same.

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