Responsibilities

• Own the technical architecture of our AI / LLM systems end to end - from framework and tooling choices through to deployment.

• Build reusable platform components: a model gateway, permission-aware RAG, agent orchestration, secure tools/MCP connectors and internal-system integrations.

• Translate security and compliance policy into technical controls for SSO/RBAC, data classification and isolation, PII, secrets, approvals and audit.

• Establish evaluation, observability and release gates covering answer and retrieval quality, safety, cost, latency, reliability and drift.

• Partner with product and business teams to turn priority problems into shipped, dependable solutions, with the right human oversight.

• Keep the architecture modular and vendor-neutral, so we are never locked to a single framework or model provider.

Qualifications and Experience

• 5+ years software engineering, with strong Python and production backend/API experience and ownership of 0-to-1 systems.

• Hands-on production experience with LangChain/LangGraph, OpenAI Agents SDK, Microsoft Agent Framework/Semantic Kernel or comparable agent SDKs, including durable workflows, structured outputs, tool calling and HITL.

• Experience building permission-aware RAG and ingestion pipelines with LlamaIndex, Haystack or LangChain and vector stores such as pgvector, Pinecone or Qdrant, including source ACLs and data isolation.

• Strong cloud-native engineering fundamentals, plus evaluation and tracing with LangSmith, Arize Phoenix, DeepEval, Ragas or equivalent, including CI/CD release gates and incident readiness.

• Hands-on experience implementing guardrails with NVIDIA NeMo Guardrails, Guardrails AI, OpenAI Guardrails or equivalent, covering prompt injection, PII/data leakage, output validation and unsafe tool use.

• Experience implementing fine-grained permissions and secure tools/connectors with OPA, OpenFGA, Cedar or equivalent, plus MCP, OAuth/service authentication, least-privilege scopes and approval gates.

• Clear communicator who turns ambiguous problems into scoped milestones, architecture decisions, runbooks and shipped, reliable software.