AI Evals Engineer - Evaluation Datasets & Ground Truth

Posted 2 days ago

prophetic technologiesDenver (CO)

Validation Engineers Software Publishers

SENIORITY

Lead

Apply

About the role

Prophetic Software is not able to sponsor employment visas now or in the future. Candidates must be authorized to work in the United States without current or future sponsorship to be considered for this role. About Prophetic: Real estate development is a multi-billion-dollar industry that has run on fragmented data, manual processes, and gut instinct for decades. Prophetic is changing that. We're building the AI-native platform that enables homebuilders, developers, and investors to find, analyze, and act on land opportunities from a single system - powered by proprietary technologies that process billions of data points across all 50 states. We are the market leader in our space, and our customers don't just use the product - they love it. We're not making teams more efficient. We're changing how they operate. Prophetic is scaling fast - demand is outpacing our ability to hire, and we're just getting started. This is a once-in-a-lifetime team: sharp, low ego, deeply collaborative, and obsessed with building the best product in the industry. Come disrupt an industry with us. Why this role exists We iterate on pipeline components constantly: prompts, models, classifier thresholds, retrieval logic, harness design. We are hiring a full-time engineer to produce, maintain, and defend the evaluation datasets that let every meaningful component in our stack be measured. You will own ground truth at Prophetic. This is not an evals-infrastructure role and it is not a labeling-operations role, although you'll touch both. Your deliverable is trusted data: for a given module, a versioned set of inputs and expected outputs, plus a written definition of what "correct" means and how confident we should be in the labels.

What you'll do:

Decide what needs to be measured, and how
Read our pipelines and system architecture, sit with product and engineering, and decompose each system into evaluable modules with explicit input → expected-output contracts.
For each module, define what "correct" means in writing: rubrics, label schemas, edge-case policies, and the tolerances that matter to the business.
Prioritize. We have more modules than you can cover in year one; you'll decide where a validation set unblocks the most iteration, with minimal direction from principal engineers.
Source the data by whatever means is cheapest and most trustworthy for that module
Production sampling: pull stratified, de-identified samples from real traffic so eval sets reflect what the system actually sees, including the long tail.
Human labeling: scope and run labeling programs, write annotation guidelines, build calibration sets, measure inter-annotator agreement, and manage vendors (including offshore labeling teams) or internal subject-matter experts. You own label quality, not just label throughput.

Synthetic / oracle-generated ground truth: where a task is tractable for a frontier model given enough compute (long context, multi-pass, tool use, self-consistency) but too expensive to run that way in production, design the oracle harness that produces labels for the cheap production path to be measured against. Then verify the oracle: calibrate its output against a human-labeled sample before anyone trusts it. Programmatic and adversarial construction: heuristic labels, templated edge cases, backtests built from past incidents ("what test would have caught this?").Make the data trustworthy over time Version every dataset; track lineage, splits, and which model/prompt versions have seen which examples. Guard against contamination and leakage (eval examples drifting into few-shot prompts, the oracle model also being the production model, etc.).Slice by customer segment, input type, and difficulty so a headline number can't hide a regression. Refresh sets as the product and traffic change; retire stale examples. Close the loop with engineering Calibrate automated graders (LLM-as-judge, similarity metrics, exact-match) against your human gold sets, and be the person who says when an automated judge is good enough to gate on. Report metrics correctly: precision/recall/F1, confusion matrices, calibration, confidence intervals, sample sizes needed to detect a given effect. Partner with engineers who own the eval harness and CI so your datasets are actually run, and with ML engineers on classifier features when the data tells you the features are the problem.

What we're looking for:

Must have
Experience building or evaluating ML or LLM-powered systems in production, in a role where output quality was your problem.
You have built evaluation or validation datasets before and can talk about one in detail: how you defined correctness, how you sourced labels, what went wrong, how you knew the labels were good.
Working fluency in ML validation fundamentals: train/validation/test discipline, stratified sampling, precision/recall trade-offs, class imbalance, calibration, inter-rater agreement, basic significance testing and power.
Understand feature engineering well enough to reason about why a classifier fails and what data would expose it.
Strong Python and SQL; comfortable pulling and reshaping data yourself.
Hands-on with LLM-based systems: prompting, structured outputs, agent/tool-use harnesses, and the specific ways they fail (non-determinism, prompt sensitivity, evaluator bias).Judgment about when LLM-as-judge is reliable and when it is not, and how to prove either.
You can read a system design, understand the business logic it encodes, and translate that into a label schema without waiting to be told.

Nice to have:

Have run a human-labeling program end to end, including vendor selection, guideline authoring, QA sampling, and cost/quality trade-offs.
Experience with eval tooling.
Experience with data labeling platforms.
Have used a frontier model as a distillation/oracle source and can articulate where that assumption breaks.
How you think
Verify the verifier. A label is a claim, not a fact, until something independent agrees with it.
Start simple: binary before graded, one judge before five, a hundred well-understood examples before ten thousand noisy ones.
You measure business outcomes, not model vibes, and you're comfortable telling a senior engineer their favorite change didn't move the number.
You'd rather own an unglamorous problem completely than a glamorous one partially.
Team & reporting
You'll report to the Chief AI Officer and work across all product/pipeline teams. Over time it is expected that you'll lead a small team of eval engineers. You'll have a generous budget for labeling vendors and oracle compute.
Why Prophetic:
The team. Wide expertise, sharp, low ego, and genuinely fun to work with. We have each other's backs and we celebrate wins together.
Real ownership and impact from day one. Our customers make million-dollar decisions on our platform. Your work matters immediately.
AI-native product and AI-native workflows. Every team - engineering, sales, operations - runs on AI tooling daily.
Founder-led transparency. Leadership shares the financials, the strategy, and the hard calls with the whole team.
High growth, high demand. The product works, customers love it, and we are hiring to keep up. No shortage of opportunity here.
Flexibility. Remote, hybrid, or in-office with a great Portland, OR headquarters.
Self-starters who thrive in ambiguity. We're a startup. If you need someone to tell you what to do every morning, this isn't the place; if you come alive solving hard problems with smart people, it is.

Benefits:

(US-based Full-time):100% medical, dental & vision insurance coverage for you; 30% coverage for dependents
Competitive salary and meaningful early-stage equity
Unlimited PTOHybrid/Remote stipend

In-office perks: snacks, drinks, coffee, ping-pong table, and more - plus cuddles from Olive, our in-office Doberman! Budget for intra-office travel2-3 annual team meetups in person How We Work: These four themes are how we think, build, and lead: Truth - Say the hard thing. We speak plainly, flag problems early, and share what's actually going on. Precision - Sign your name to it. Our customers make million-dollar decisions on our data. We get the details right. Velocity - Speed compounds. Waiting doesn't. We ship daily and launch features every two to three weeks. Ownership - Own it, then solve it. We own outcomes, not tasks. If something's broken, we fix it. EEO:Prophetic Software is an equal opportunity employer. We are committed to creating an inclusive environment for all employees and applicants, and we make employment decisions without regard to race, color, religion, sex (including pregnancy, childbirth, or related medical conditions), gender identity or expression, sexual orientation, national origin, age, disability, genetic information, veteran or military status, marital status, or any other characteristic protected under applicable federal, state, or local law. We actively seek to build a diverse team and encourage candidates from underrepresented backgrounds to apply. Reasonable accommodations are available upon request for qualified individuals with disabilities throughout the application and employment process.

Before you apply

Applying takes about a minute. These four things decide how fast it moves after that.

Your profile is current

It's what we read first. Occupations, seniority and locations matter more than a long history.

Two examples you can talk through

Not a portfolio — just two pieces of work where you can explain the decisions and what you'd change.

A number in mind

What you're on now and what would make you move. We negotiate better when we know both.

Your notice period

Employers plan around it, and it's the question that stalls offers most often.

Once you apply, someone reads it and calls you before anything reaches the employer — usually within two working days.