Backend Infrastructure & Agentic AI Platforms Software Development Engineer, Senior

Booz Allen Hamilton•Washington, DC

1d•$86,800 - $198,000•Onsite

About The Position

To achieve an organization’s mission, leaders need strong team members who can build the next generation of agentic AI to transform how clients accelerate research, makes decisions, and ships products at scale. That is why we need you, an experienced Software Development Engineer who can operate at a system-of-systems level to support clients in advancing AI-enabled systems within an R&D environment. As part of our team, you'll serve as a Software Development Engineer to the Advanced Research Projects Agency for Health (ARPA-H). ARPA-H has a small team that is building the next generation of agentic AI to transform how the agency accelerates research, makes decisions, and ships products at scale. The team will evolve ARPA-H's production AI assistant into an ecosystem of autonomous, multi-agent systems. You'll serve as a Software Development Engineer where you will own the backend infrastructure, while also being a first-principles builder of the agentic AI systems that run on top of it. On a lean team, infra and AI are not separate concerns. You will own both, and you will treat production reliability, token economics, security, and observability as non-negotiable from day one.

Requirements

7+ years of experience with software engineering, including building and operating production systems
Experience in high-velocity environments where you owned and shipped complex products end-to-end
Experience with at least 2 backend languages, including Python
Experience being on-call, debugging incidents, and writing the postmortem
Experience with Microsoft Azure, including Azure Functions, API Management, Container Apps, and Azure OpenAI Service
Experience with containerization, CI/CD, and infrastructure as code
Knowledge of authentication and identity systems, such as OAuth2, OIDC, or Azure Entra ID
Knowledge of modern backend frameworks and async patterns, distributed systems, APIs, data pipelines, and software design patterns
Ability to own production systems
Bachelor's degree in Computer Science or Software Engineering

Nice To Haves

Experience building and operating MCP servers in production, including tool registration, versioning, and hosting on Azure Functions or equivalent serverless infrastructure
Experience implementing A2A communication patterns and multi-agent orchestration frameworks
Experience building on top of LLMs in production, including tool-calling, RAG, multi-step reasoning, multi-model routing, and context window management
Experience in token economics, including cost-per-query, context budgets, and prompt efficiency as first-class engineering concerns
Experience managing multi-provider LLM integrations including rate limits, fallback routing, and API versioning
Experience in security-conscious engineering in regulated or government environments, including tracking record in startup or early-stage environments, including 0-to-1 product building
Experience in big tech building customer-facing platforms or developer infrastructure at scale
Knowledge of vector databases, embedding pipelines, and semantic search infrastructure
Ability to be comfortable with ambiguity and a high sense of urgency, be a self-starter, operate within a fast-paced environment, multi-task, and handle multiple priorities
Possession of excellent oral and written communication skills

Responsibilities

Support backend infrastructure, agentic AI and protocol infrastructure, observability and production quality, and engineering excellence
Own the end-to-end backend infrastructure on Microsoft Azure such as Azure Functions, Azure API Management, Azure Container Apps, and Azure OpenAI Service
Own the data layer such as storage, retrieval pipelines, vector databases, and document indexing that power GRACE's internal knowledge search
Own authentication and identity integration, including ARPA-H Entra ID and application-level access control, implement and maintain infrastructure as code for all environments, and no manual snowflakes
Own CI/CD pipelines, deployment automation, and release processes including canary and gradual rollouts, own monitoring, alerting, logging, distributed tracing, SLOs, and incident response runbooks, and manage secrets, API keys, and credential rotation across all integrations with external providers
Own cost and token economics across all LLM providers and track spend, analyze budgets, build guardrails, and optimize for cost-per-query without sacrificing quality
Own the backend implementation of MCP, including MCP server hosting, tool registration, versioning, and lifecycle management on Azure and implement and evolve A2A communication patterns, enabling GRACE agents to interoperate with each other and with external agent systems
Design and maintain LLM orchestration, routing, and multi-model switching infrastructure across OpenAI GPT, Anthropic Claude, and Google Gemini families and build and operate RAG pipelines, including document ingestion, chunking, embedding, and semantic search
Implement robust fallback, retry, circuit-breaker, and graceful degradation patterns for all AI service dependencies and own tool-calling infrastructure, including registration, execution, error handling, and observability for all GRACE tools
Build and maintain end-to-end observability for agentic workflows, including latency, throughput, error rates, token usage, and LLM quality metrics and implement LLM evaluation pipelines including safety checks, regression monitoring, and grounding assessment
Define and enforce system-level SLOs for availability, response time, and tool call reliability and own alerting and on-call runbooks
Establish and improve coding standards, design review processes, and testing practices and ensure strong privacy, security, and compliance in all systems, integrations, and data handling
Communicate technical decisions clearly, in writing and in conversation, to both engineers and non-engineers
Work backward from the user and understand the problem being solved before proposing a solution