Software Engineer, AI Agent Platform

ChampionAI is QAD | Redzone agentic platform, purpose-built for manufacturing and utilized by the various business units within QAD | Redzone. This engineer joins the core platform team with three main areas of focus: building new capabilities into the platform, helping Applied AI teams at partner business units build and ship Champions correctly, and directly building Champion agents for business-unit use cases when needed. You will work in the same codebase as the Applied AI engineers you support, which keeps the enablement work practical and the platform work focused on real problems. Key Responsibilities Platform Development Build and ship features across the Champion platform repositories Improve developer experience: tooling, scaffolding, internal documentation, and onboarding paths for Applied AI engineers. Maintain and evolve the MCP tool server and agent infrastructure that BU teams depend on. Identify and address friction points that slow down Champion development or deployment Applied AI Enablement Support BU Applied AI engineers in building, deploying, and operating Champions correctly Review agent implementations for prompt quality, scope enforcement, auth configuration, and deployment setup Contribute to internal engineering guides and skill documentation Pair with BU engineers on first K8S manifest creation, database registration, and LaunchDarkly prompt rollout Agent Development Design and build Champion agents for BU use-cases when the platform team is directly engaged Write and iterate on system prompts, tool bindings, and context injection for production agents Register agents in Champion Server and configure LaunchDarkly-gated prompt rollout across environments Engineering Practices We follow trunk-based development with PR-gated merges to main. Engineers are expected to: Write tests before implementing. TDD is the expectation, not a nice-to-have. Keep PRs small and focused; use feature flags to ship partial work incrementally Follow conventional commit format (feat:, fix:, etc.) Be mindful of API and schema backward-compatibility; prefer additive changes over breaking ones Review your own diff before requesting review Core Requirements Python: Proficient in async Python, Pydantic, type hints, and FastAPI. Experience with the Strands agent SDK or a comparable agentic framework. Testing is non-optional: candidates should be comfortable with pytest, pytest- asyncio, and DeepEval for agent-specific evaluation. We test agent behavior, not just unit logic. Prompt Engineering: Able to write and iterate on production system prompts: XML-structured, scope-enforced, with tool descriptions that guide LLM delegation reliably. Model Context Protocol (MCP): Solid understanding of MCP and hands-on experience building or consuming MCP servers. Familiarity with Agent-to-Agent (A2A) protocol is a strong plus. Agent Patterns: Familiar with multi-agent architectures and orchestration patterns beyond basic ReAct: supervisor/subagent delegation, parallel tool use, handoffs, and context management across agent boundaries. Agent Evals: Able to design and run evaluation suites for agent behavior: correctness checks, scope enforcement tests, regression coverage, and systematic prompt iteration based on eval results. Docker: Comfortable authoring Dockerfiles, multi-stage builds, and local dev environments via docker-compose / make. PostgreSQL: Comfortable with templated INSERT/SELECT, foreign key relationships, and reading an ER diagram. Git / Semantic Versioning: Follows conventional commit format (feat:, fix:) and a PR-based trunk workflow. Nice to Have Kubernetes: Working knowledge of Deployments, Services, ConfigMaps, and ServiceAccounts. Able to read and adapt K8s manifests and use kubectl for basic troubleshooting. AWS: Working knowledge of ECR, EKS, IAM, and Bedrock (inference layer). OAuth2 / OIDC: Conceptual understanding of Authorization Code Flow with PKCE, token exchange, and agent auth delegation. Terraform: Beginner-level familiarity; able to make targeted changes to existing modules and interpret a plan diff. LaunchDarkly: Experience managing feature flags or AI config overrides for environment-gated rollout. Document Intelligence / OCR: AWS Textract or comparable pipeline experience for use cases involving structured document extraction. AI-Native Development We run an AI-native SDLC. AI-assisted tools (Claude Code, Cursor) are part of the standard engineering workflow at every stage: planning, implementation, review, and documentation. This is not optional or supplemental; it is how we work. Candidates should be comfortable with AI-assisted development and willing to invest in getting good at it. Use Claude Code and Cursor actively across planning, implementation, and code review Know when AI output is wrong and push back on it. Help drive adoption across the team by sharing what works What This Role Is Not This is not an ML research or data science role. The platform team owns cluster networkin…