AI Engineer

We're hiring an AI Engineer to own the intelligence layer of our platform. You'll design and build extraction pipelines, model orchestration, prompt architectures, and evaluation frameworks that turn messy insurance documents into structured, underwriting-ready data. This is a production-focused role where you'll ship AI systems that process real broker submissions, extract data from ACORD forms and policy documents, and feed structured outputs into downstream underwriting workflows. You should be comfortable with the full lifecycle from prompt design to deployment, monitoring, and iteration based on production feedback.

Responsibilities:

• Design, build, and maintain production LLM extraction pipelines that process insurance documents such as ACORD 125/140, loss runs, supplemental applications, SOVs, and SPAs into structured data.
• Own the multi-pass extraction architecture end-to-end, including OCR, line-of-business detection, line-specific structured extraction, confidence scoring, and fallback logic.
• Write, version, and optimize extraction prompts across insurance lines of business, including General Liability, Commercial Property, Workers' Compensation, Professional Indemnity, and W& I.
• Build and maintain evaluation frameworks, including ground truth datasets, extraction accuracy benchmarking, regression testing, and automated quality scoring.
• Design feedback loops that improve extraction quality over time using underwriter corrections, document template recognition, and triage calibration.
• Implement model routing and fallback strategies for cost optimization, latency management, provider failover, and model selection.
• Build document classification and triage systems that identify document types, lines of business, and routing priorities from incoming submissions.
• Own LLM observability, including prompt tracing, cost tracking, latency monitoring, A/B testing of prompt variants, and debugging extraction failures.
• Collaborate with product and engineering to define extraction schemas, field mappings, and output contracts consumed by downstream systems.
• Evaluate and integrate new models, techniques, and tools as the LLM landscape evolves, including vision models, structured output modes, fine-tuning strategies, and agentic patterns.

Requirements:

• 4+ years of software engineering experience with strong Python proficiency and at least 2 years of hands-on experience building AI/ML-powered production systems.
• Deep practical experience with LLM APIs in production, including prompt engineering, output parsing, structured extraction, token management, error handling, and retry strategies.
• Experience building document understanding or information extraction pipelines, including OCR, PDF processing, form extraction, or NLP-based data extraction at scale.
• Strong understanding of evaluation methodology, including measuring extraction quality, building ground truth datasets, and setting up automated regression testing for AI outputs.
• Hands-on experience with at least one LLM observability or evaluation platform such as Langfuse, LangSmith, Braintrust, Weights and Biases, or equivalent.
• Production experience with AWS or an equivalent cloud platform, including deploying and monitoring AI pipelines.
• Daily use of AI tools in your engineering workflow, such as Claude Code, Copilot, Cursor, or equivalents, for coding, debugging, and research.
• Strong ownership mindset with the ability to identify extraction quality issues, propose solutions, build them, and measure impact.
• Clear communication skills with the ability to explain model trade-offs, extraction accuracy metrics, and pipeline architecture to non-specialists.

Nice To Have:

• Experience with vision-language models for document understanding, such as GPT-4V, Gemini Pro Vision, Claude with Vision, DocTR, or LayoutLM.
• Familiarity with insurance documents and data standards such as ACORD forms, policy administration schemas, SOVs, and loss runs.
• Experience with workflow orchestration frameworks such as Temporal, Airflow, or Prefect for managing multi-step AI pipelines.
• Hands-on work with vector databases and embedding-based retrieval such as pgvector, Pinecone, or Weaviate for RAG or document similarity.
• Experience with fine-tuning or distillation techniques for domain-specific model performance.
• Background in building multi-agent or agentic AI systems, including tool use, chain-of-thought orchestration, and agent-based workflows.
• Contributions to open-source AI/ML projects or published technical writing demonstrating technical depth.
• Prior experience in insurance, fintech, legaltech, or other document-heavy regulated domains.