All roles
Applied AI
Senior NLP Engineer, Legal Reasoning
Delhi, India (on-site, East Delhi) · Full-time
Build the models that read adjuster diary notes, police narratives and fire origin and cause reports, and decide whether a third party caused the loss. You will turn messy claim text into fault findings that subrogation counsel can check line by line, and you will own the evaluation work behind our 92% adverse liability scoring accuracy.
What you will do
- Design extraction and reasoning pipelines that identify parties, actions, negation and fault allocation in claim narratives written in adjuster shorthand.
- Build document understanding for scanned police crash reports, origin and cause reports and repair estimates, including OCR error handling and table extraction.
- Turn state fault rules and liability theories from our legal research team into structured reasoning steps the model must show with every score.
- Own evaluation: counsel-labeled test sets, error analysis by line of business and state, and regression gates before each release.
- Work within our zero-training rule: models learn from synthetic, public and licensed text annotated in-house, never from customer claim files.
- Partner with engineering on latency and cost so 50,000-file closed-claim sweeps finish inside 48 hours.
What you bring
- 5+ years building NLP or ML systems that run in production, with at least 2 on transformer or large language model applications.
- Strong Python and PyTorch, plus experience with retrieval, structured output and evaluation tooling for LLMs.
- Hands-on information extraction from noisy, domain-specific text such as clinical, legal, financial or insurance documents.
- Experience designing labeled datasets and annotation guidelines with subject-matter experts.
- Clear written explanations of model behavior for readers outside ML, including attorneys and claims leaders.
- A degree in computer science, computational linguistics or a related field, or equivalent experience.
Nice to have
- Exposure to US tort concepts such as comparative negligence, product liability or statutes of limitation.
- Work on OCR, layout-aware models or form understanding.
- Publications or open-source work in information extraction or legal NLP.