The RoleCCC Intelligent Solutions Inc. (CCC) is a leading cloud platform for the multi-trillion-dollar insurance economy, creating intelligent experiences for insurers, repairers, automakers, part suppliers, and more. At CCC, we’re making life just work by empowering more than 35,000 businesses with industry-leading technology to get drivers back on the road and to health quickly and seamlessly. We’re pushing boundaries with innovative AI solutions that simplify and enhance the claims and repair journey. Through purposeful innovation and the strength of its connections, CCC technologies empower the people and industry relied upon to keep lives moving forward when it matters most. Learn more about CCC at www.cccis.com.
The Role:
You will build and ship LLM-powered features and agentic workflows for our customers. These features automate the routine parts of a claim and guide customers on the rest, so they have to earn that trust with measured accuracy, not a convincing demo. 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 design, security, and hard debugging. You own features end to end.
Key Responsibilities:- Ship LLM-powered features and agentic workflows in Python and TypeScript, from a problem scoped with product managers and data scientists through production traffic, and help set the metrics you will be measured against.
- Write the evals, guardrails, and feedback loops that make your features trustworthy, and hold them to explicit token and latency budgets.
- Use agentic coding tools every day, and own the review of what they produce: catch the correctness, security, and design problems in generated code before you open the pull request.
- Maintain and extend the NLP pipelines that turn unstructured claim data into structured data customers can act on, blending semantic search, classical NLP, and LLMs.
- Build and operate durable, event-driven services in which agents run as steps inside deterministic workflows, with checkpointed model outputs so a retry does not re-invoke a completed LLM step.
- Trace every model call and agent step so a recommendation can be reconstructed and audited, with citations back to the source documents, and wire that tracing into your alerts.
- Model and query claims data across relational, document, and graph stores.
- Design privacy and security in from the start: tenant isolation, least-privilege access for agents and their tools, and a clear rule for what data may reach a model provider.
- Write a technical design document before any significant change, and use it to get feedback and buy-in from the team before you build.
- Take part in a shared on-call rotation for the services you build.
Requirements:- 3+ years of professional software engineering experience, including at least one LLM-powered feature you shipped and then operated in production.
- Python and TypeScript: typed, tested, async code in both, and real depth in one.
- Evals you have designed for LLM output, plus guardrails you have shipped.
- 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.
- Pipelines that blend LLMs with semantic search and classical NLP: structured output, tool calling, embedding-based retrieval, and a case where a classical technique beat an LLM.
- Services built on message- or event-driven architectures, including long-running workflows with retries, idempotent handlers, and event-sourced state, plus an LLM or agent step you ran inside one.
- 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.
- Backend services built and operated on Azure or AWS, and exposure to a durable workflow engine such as Temporal or the Durable Task Scheduler.
- Technical design documents you have written to get feedback and buy-in from a team before building, and that the team then built to.
- Clear writing and discussion for technical and non-technical audiences, and code review you can give as well as take.
Even Better if You Have:
- React and TypeScript experience, so you can deliver a full-stack feature end to end. Chat, streaming, and citation-heavy AI experiences especially.
- Experience with CI/CD pipelines and infrastructure as code such as Terraform or Bicep.
- Experience with agent frameworks, LangGraph 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, or what-if simulations that replay historical claims through a proposed change before it ships.
- Exposure to the insurance, claims, or automotive repair domains.
What Success Looks Like:
- In your first six months you ship an LLM-powered feature, its evals run in CI, and you can name the metric it moved.
- When a model migration or prompt change causes a regression, the evals you wrote catch it before a customer does.
- You take ownership from concept to production with light direction, asking the right questions early.
- You care that things hold up: the services you build survive retries, partial failures, and model migrations without incidents a customer can see.
- When something goes wrong in production, the trace tells you which step and which model call, and you can show an auditor the same thing.
- Privacy and security reviews of your features find nothing that was not already designed in.
- You enjoy the craft, and it shows: engineers pick up your patterns without being told to.
- You enjoy the back-and-forth that gets a feature scoped, 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.
