Data Engineer (AI Infrastructure)

Location: Irvine, CA | On-site || Department: Technology || Employment Type: Full-time

About the role

You'll build the data systems and services behind our AI platform. You'll turn complex financial and case documents into structured, trustworthy data for our AI, analytics, and operational systems, and help build the real-time services the organization depends on. You'll join a team that already has a working platform with design docs, tests, and CI, and help shape where it goes next. The work is hands-on, from design through deployment.

What you'll do

Data

Build batch and real-time pipelines that bring in data from documents, APIs, CRM systems, and internal services.
Build document-processing systems on OCR and AI models that stay accurate, reliable, and affordable as volume grows.
Design data models across relational databases, document stores, and data lake storage, and change them safely as the product evolves.
Give AI/ML teams clean, structured, accessible datasets, including training data.
Build data quality checks and monitoring so bad data is caught before anyone relies on it.

Platform and services

Build and run APIs and background services, and own how they behave when the systems they depend on are slow, failing, or rate-limited.
Maintain integrations with our CRM and other third-party systems, and fix them when they break.
Help build real-time sales tools where speed and correctness both matter, such as lead routing and live caller lookup.
Contribute to secure, controlled data exports to external partners.
Maintain older services and migrate them to our primary cloud where it makes sense.
Support the internal web tools our case and sales staff use.

Engineering practice

Write design docs for your own work and contribute to architecture decisions.
Work in our infrastructure-as-code and CI/CD setup, changing it as your work requires.
Protect sensitive personal and financial data in code, logs, configuration, and third-party integrations.
Mentor junior engineers and help improve our standards for code quality, CI, and releases, including guardrails for AI coding agents.

What You Bring

We care more about fundamentals than any particular stack. In each area below, we're looking for someone who can explain why things work the way they do, not just name the tools.

Experience. 2-3 years building production backend or data systems, or equivalent depth: you've owned production systems end to end, made the design decisions, and handled the incidents. A degree in Computer Science, Engineering, or a related field, or equivalent practical experience.
Software engineering. You treat engineering as more than programming. You frame the problem, weigh the options, write down the trade-offs, and choose the simplest design that works. You ship code that's tested, observable, and easy to change. You measure before you optimize, and you own what you build after it ships.
Languages. Deep experience in at least one backend language. We work mostly in Python.
Databases. Strong SQL and data modeling across relational and document databases, and a clear sense of when to use which. You understand indexes and query plans, transactions and isolation, schema changes that don't break running systems, and keeping data consistent when it lives in more than one store.
Data pipelines. You've built pipelines over structured and unstructured data (JSON, PDFs) that are safe to rerun and backfill, and you understand how to model data for analytics.
AI systems. You've run LLM or OCR workloads in production and know how they behave: output that varies from run to run, structured output that needs validating, quality you have to measure, and latency and cost that grow with volume.
Model deployments. You understand how models are deployed and served: capacity and quotas, pinning versions and evaluating new ones before switching, rolling back, and falling back when a model or region is unavailable.
Rate limits. You understand rate limiting from both sides: how a client should handle a 429, how a service protects itself with quotas and backpressure, and the various 429 errors that can occur for LLM deployments.