Note: The job is a remote job and is open to candidates in USA. Cypress HCM is building a Centralized Agentic AI Framework to integrate safe and governed AI agents into GitLab workflows. The role involves hands-on engineering to extend the platform, focusing on agent orchestration, CI/CD integration, and implementing security measures.
Responsibilities
- Build agent orchestration — implement Router Lambdas, SQS-based queueing, and Bedrock Agent invocation
- Integrate with GitLab CI/CD — design .gitlab-ci.yml patterns where agent invocations run as pipeline stages, consume branch/diff/test context, emit artefacts, and gate downstream stages on agent pass/fail signals
- Develop shared Action Groups — build the AWS Lambda-backed tools
- Design Knowledge Bases — iUtilise AWS Bedrock Knowledge Base / OpenSearch / S3 so agents reason over shared organizational context, not isolated prompts
- Implement Bedrock Guardrails — input/output filters, sensitive-data scrubbing, content/word filters, and per-agent permission boundaries enforced by design
- Implement TokenOps controls — model tiering and routing via an LLM gateway, semantic caching, context-window management
- Instrument everything in AWS — CloudWatch dashboards, audit trails, trace logging of every agent invocation (input, output, decision, tokens, cost) for compliance and debugging
Skills
- 3–7 years professional experience
- Strong Python, AWS SDK Python for building production Lambda functions and event-driven services
- Solid AWS experience: Lambda, API Gateway, SQS, EventBridge, IAM, Secrets Manager, CloudWatch
- Amazon Bedrock specifically: Agents, Action Groups, Knowledge Bases, Guardrails
- Hands-on experience integrating with LLM APIs — Bedrock, OpenAI, Anthropic, or similar
- Familiarity with CI/CD platforms — GitLab CI strongly
- A security mindset: least-privilege IAM, secrets handling, input validation, awareness of prompt injection and data-leak risks in LLM workflows
- Comfortable with observability — structured logging, metrics, tracing — and writing code that's debuggable in production
- Vector search / RAG: OpenSearch, embeddings, retrieval evaluation
- FinOps or TokenOps: cost attribution, model routing, semantic caching, batch inference
- Experience building developer platforms or internal tooling — you've shipped something other engineers depend on daily
- Familiarity with agent frameworks (Bedrock AgentCore) and their tradeoffs
Company Overview