Gen AI Developer / Lead / Architect

About the role

You build the AI agents at the heart of **M-agentic — Agentic Process Studio**. Working within the agent squad, you implement and test the agents that automate a real **Health Insurance Claim Adjudication** process on **AWS AgentCore + Amazon Bedrock (Claude)** — writing the prompts, tools, guardrails and integrations that make each agent reliable, safe and auditable.

Key responsibilities

• Implement worker and application agents on **AWS AgentCore / Bedrock**: prompts, tool definitions, control flow and hand-offs between steps.

• Engineer and iterate prompts for the Studio’s AI features (intake parsing, gap analysis, agent-blueprint generation) and contribute test cases to the **eval harness**.

• Integrate enterprise systems as governed tools (**API connectors / MCP tools**) for the six claims systems (Availity, HealthEdge, Optum, Shift, Zelis, InterQual) using sandbox adapters.

• Wire safety and compliance into every agent: **Bedrock Guardrails**, PHI masking, and the immutable audit trail.

• Write clean, tested **Python**; participate in code review and pair with the Lead and peers.

• Help implement the human-in-the-loop gates (clinical review, SIU) and the Forge deploy path.

• Debug agent behaviour using traces, logs and evals; tune model choice and cost per step.

Required qualifications

• 2+ years in software or ML engineering with strong **Python**.

• Practical experience with **LLM applications** — prompt engineering, tool/function calling, and at least one agent or RAG project (professional, open-source or substantial personal).

• Working knowledge of **AWS** (Lambda, IAM, S3, and event-driven basics).

• Solid fundamentals: version control, testing, APIs, JSON/data handling.

• Curious, detail-oriented, and comfortable with fast iteration and measurement.

Preferred

• Exposure to **Amazon Bedrock**, **AWS AgentCore**, Bedrock Agents, LangGraph/LangChain, or **MCP**.

• Experience integrating third-party APIs or building connectors.

• Interest in AI safety, guardrails, evals, or regulated-data handling.

What success looks like (first 8 weeks)

• Owns one or more worker agents end-to-end — built, tested and running in the pipeline.

• Contributes reliable prompts and eval cases that raise output quality measurably.

• Ships steadily within the squad’s cadence with clean, reviewed code