6+ years of software engineering experience, including architecting and shipping LLM-powered systems.
Proficient in Python and TypeScript, with experience in setting team conventions and performance profiling.
Experience in establishing evals and guardrails strategies for product areas.
Demonstrated ability to constrain agentic coding tools and verify outputs effectively.
Designed NLP pipelines integrating LLMs, semantic search, and classical NLP techniques.
Experience with durable, event-driven systems and deterministic orchestration.
Familiarity with data modeling across various database types, including PostgreSQL and MongoDB.
Responsibilities
Own the delivery of LLM-powered systems from problem scoping to production operation.
Set evals and guardrails strategy for releases and monitor production signals.
Enhance team workflows with agentic coding tools and maintain code review standards.
Design NLP pipelines to convert unstructured claim data into actionable structured data.
Build and operate multi-tenant systems with deterministic workflows and defined failure modes.
Manage tracing and audit processes for model calls and agent steps.
Establish privacy and security baselines for all systems in your area.
Benefits
Flexible work arrangements to promote work-life balance.
Opportunities for professional development and continuous learning.
Collaborative and inclusive company culture.
Access to cutting-edge technology and tools.
Participation in a shared on-call rotation for system ownership.
Full Job Description
The Role The Role: You will own LLM-powered systems and agentic workflows for our customers. These systems automate the routine parts of a claim and guide customers on the rest, so they have to earn that trust with a measured accuracy you can defend as models change and usage grows. Agents run as steps inside deterministic workflows, and privacy and security are design baselines, not review gates. Agentic coding tools are how you move fast. What you add is what they are worst at: system architecture, code quality, security, and the debugging nobody else gets to the bottom of. You own a product area end to end and define the eval bar it ships against.
Key Responsibilities:
Own the delivery of LLM-powered systems in Python and TypeScript, from an ambiguous customer problem, scoped with product managers and data scientists, through production operation.
Set the evals and guardrails strategy for your area: what gates a release, what production signals the team watches, and how a change is simulated against historical claims before it ships.
Raise the bar for how the team works with agentic coding tools: sharpen the specifications, instruction files, hooks, and permission policies they run under, and hold generated code to the same review standard before it merges.
Design the NLP pipelines that turn unstructured claim data into structured data customers can act on, choosing where semantic search, classical NLP, or an LLM fits.
Build and operate durable, event-driven, multi-tenant systems in which agents run as steps inside deterministic workflows, designing their failure modes, replay semantics, and cost profile before they show up under load.
Own tracing and audit for your area: every model call and agent step recorded, so any recommendation can be reconstructed and cited back to its sources by on-call and auditors alike.
Set the privacy and security baseline every system in your area starts from: tenant isolation, least-privilege access for agents and their tools, redaction, retention, and what data may reach a model provider.
Author the design documents for complex systems to get feedback and buy-in, review other engineers' designs and code with the same care, and mentor the engineers in your area.
Take part in a shared on-call rotation for the systems you own.
Requirements:
6+ years of professional software engineering experience, including an LLM-powered system you architected, shipped to external customers, and carried through at least one model or prompt migration.
Python and TypeScript at the level of setting a team's conventions: the async model, typing strictness, and test strategy for a shared codebase, plus a performance problem you profiled and fixed.
An evals and guardrails strategy you set for a product area others built in: where human approval was required, and how production corrections re-entered the evals.
A clear account of how you constrain an agentic coding tool's context and verify its output, from daily use of Claude Code, Cursor, GitHub Copilot, or similar on production work, plus changes you turned back, and why.
Pipelines you designed that blend LLMs, semantic search, and classical NLP, including a retrieval design you can defend and a case where a classical technique beat an LLM on cost, latency, or reliability.
Durable, event-driven systems you designed with deterministic orchestration around non-deterministic model steps: compensating actions, idempotency, backpressure, and event-sourced replay.
Data modeling and development across relational, document, and graph databases such as PostgreSQL, MongoDB, and Neo4j, including when each type fits and the partition-key, consistency, and tenant-isolation calls you made.
Systems architected and operated on Azure or AWS, with CI/CD and infrastructure as code, and hands-on time with a durable workflow engine such as Temporal or the Durable Task Scheduler.
Technical design documents you have authored for systems that span more than one team, used to get feedback and buy-in before building, and that those teams then built to.
Clear writing and discussion for technical and non-technical audiences: drawing out a customer's real problem and explaining the trade-offs back to them.
Even Better if You Have:
React and TypeScript experience, so you can deliver a full-stack feature end to end, especially chat, streaming, and citation-heavy AI experiences.
Experience with agent frameworks, LangGraph, PydanticAI, and the Microsoft Agent Framework among them, and with the ways agents reach their tools: the Model Context Protocol (MCP), the Agent2Agent (A2A) protocol, command-line tools, and microservices.
Experience with LLM observability and evaluation tooling such as LangSmith, Arize, or Braintrust, and with OpenTelemetry-based tracing.
Experience designing A/B tests for AI features.
Exposure to the insurance, claims, or automotive repair domains.
What Success Looks Like:
In your first six months, every model change in your area is gated by evals you designed, regressions are caught before a customer sees them, and you can show the accuracy number held or moved.
You are trusted with the high-stakes, ambiguous problems, and you turn them into a roadmap, a shipped system, and a number you can put in front of a customer.
You are energized by what agentic coding tools make possible. Your team ships more than its headcount suggests, and your specifications and reviews are why the output survives production.
You have a passion for systems that last: the ones you design have their failure modes understood in advance, and when one fails anyway, the trace says exactly where.
Privacy and security reviews of systems in your area find nothing that was not already designed in.
Standards you set for evals, reviews, and design documents outlive the project they started on.
Engineers you mentor ship LLM-powered features with evals and guardrails in place without your direct involvement.
You enjoy getting a product manager, a data scientist, and a customer's team into one room and coming out with a single agreed scope, and you can argue a position without needing to win it.
You embrace our ambitious, collaborative, and empathetic values, and you like problems where the answer has to be measured rather than argued.
Interview Policy & Privacy Notice:
A video interview is required for this position. Video interviews are transcribed. Transcriptions are retained and may be reviewed by CCC and our recruiters. Candidates are not permitted to use generative AI or automated assistance during the interviews unless explicitly allowed by the interview team for a specific exercise. Our Job Applicant Privacy Notice is available HERE.
