NLP Scientist

GTT•San Francisco, CA

1d•Hybrid

About The Position

Our client is seeking an NLP Scientist to build evidence-grounded claim verification capabilities for regulated content review workflows. This role will focus on retrieval, claim decomposition, entailment, evidence attribution, citation quality, abstention logic, and human-in-the-loop decision support. The solution will assist expert reviewers by identifying whether generated claims are supported, contradicted, qualified, or insufficiently evidenced. Preference is for Bay Area–based candidates with the ability to work onsite at least 3 days per week. Otherwise, candidates must provide 4–6 hours of daily Pacific Time overlap. The team's preference is for candidates to be based on the West Coast, ideally in San Francisco. Remote candidates will also be considered, provided they can work Pacific Time hours.

Requirements

Strong Python production engineering with modern NLP frameworks.
Demonstrated experience in evidence-grounded NLP, retrieval, NLI/entailment, claim verification, and citation/evidence evaluation.
Experience with scientific, technical, legal, regulatory, or other high-stakes source material.
Experience building evaluation datasets and measuring whether outputs are genuinely supported by cited sources.
Experience with vector and lexical retrieval, model APIs, production evaluation infrastructure, and traceable decision systems.
Ability to work with legal, regulatory, compliance, or expert-review stakeholders.

Nice To Haves

Experience combining deterministic rules with model judgment.
Knowledge graphs connecting claims, evidence, references, products, or indications.
Experience in regulated or high-stakes review environments such as legal, financial compliance, scientific publishing, fact-checking, or life sciences.
Familiarity with study design, statistical evidence, and citation practices.

Responsibilities

Build NLP systems that verify generated claims against approved evidence sources.
Develop hybrid retrieval, natural-language inference, entailment, claim decomposition, and evidence attribution capabilities.
Design systems that return support/contradiction/insufficient-evidence decisions with traceable citations.
Build expert-labeled evaluation datasets, annotation guidelines, and inter-annotator agreement processes.
Measure false approvals, false rejections, abstention quality, citation quality, and evidence support.
Design human-in-the-loop workflows, confidence thresholds, escalation paths, and safe failure behavior.
Ensure prior decisions can be reconstructed using model version, evidence set, and reviewer action.