Founding ML Engineer

The Role

You’ll own the company’s compliance and intelligence layer end to end and work with the CTO on making AI-led interactions accurate, grounded, and safe. You’ll write code regularly, decide what is a model problem and what isn’t, and build the measurement practice the company relies on.

You are the first dedicated ML hire. You inherit a running production pipeline and real design-partner data, not a blank repository. Our test suite fails if an unearned compliance claim, an ungrounded model output, or an unmeasured accuracy number ships. You’ll hire the ML engineers who follow and set the technical bar for them.

What You’ll Own

Compliance, hallucination, and scoring architecture. Turn customer and regulatory rubrics into measurable detectors for disclosures, violations, grounding and hallucination failures, risk, escalation, and quality. Choose the right mix of rules, classifiers, judges, and fallbacks, and set confidence and promotion thresholds against explicit accuracy, cost, and risk trade-offs.
Evaluation and the data flywheel. Own annotation guidelines, adjudication, offline and online evaluation, regression gates, and per-customer reporting across compliance, grounding, and quality. Turn QA reviews, disputes, and human-versus-machine agreement into training data, and use synthetic data only where it improves downstream accuracy.
Rubric onboarding. Build the tooling, prompts, datasets, and active-learning loops that turn a new customer’s framework into a measured production detector quickly.
Speech quality. Make noisy transcripts reliably scorable today, and choose and evaluate ASR and diarization as raw-audio ingestion comes online.
Production economics and observability. Own cost and latency per interaction, routing across model tiers, and continuous monitoring for drift in compliance detection, grounding and hallucination detection, and scoring quality.

You’ll work directly with design partners’ QA and compliance teams, where most signal about model errors originates, and you’ll take part in the enterprise security and AI-governance reviews that gate each deal.

What We’re Looking For

Staff or principal depth in applied ML, NLP, or LLM systems. You’ve owned models other teams depended on, set technical direction, and been accountable when the numbers were wrong.
Production ML or LLM experience where accuracy was measured, not assumed. You’ve built evaluation pipelines, constructed ground truth from inconsistent human labels, and made explicit decisions about acceptable error rates.
Startup experience is required. You’ve shipped with broad ownership, ambiguity, and little supporting infrastructure.
Range in model selection: prompting, retrieval, fine-tuning, distillation, classical classifiers, and recognizing when the answer is a regex. You can explain the accuracy, latency, and cost trade-offs.
Hands-on production engineering. You’ve trained and debugged models yourself, shipped maintained Python services, and worked with messy unstructured data and the annotation operations that make it usable.
Production judgment on inference cost, latency, and sensitive or regulated data, including the enterprise security reviews that come with it.
You want to build the ML team while staying hands-on, and you use coding agents for leverage without lowering the quality bar.

Bonus

Speech and audio in production: ASR, diarization, streaming transcription, or real-time latency budgets.
Custom NLP or scoring-model development, or LLM-as-judge systems validated against human ground truth.
Weak supervision, programmatic labeling, or active learning in production.
Domain experience in trust and safety, fraud and risk, contact-center QA, or compliance and GRC.