Principal Product Engineer - Evinova

AstraZeneca; Gaithersburg (MD); On-site; USD 180,000 - 230,000; Full time

Job summary

AstraZeneca is seeking a pragmatic builder-architect to lead hands-on coding and architectural direction for AI-powered features. You will own end-to-end delivery, from idea through production, embedding with a product team and potentially mentoring engineers.

You will balance rapid shipping with durable design, ensuring scalable architecture and strong engineering judgment for long-term product health. This role is based in Gaithersburg, MD.

Qualifications

• Bachelor's degree in a software engineering field.
• 8+ years of software engineering experience at senior levels.
• Production experience shipping AI-powered features (agent systems, RAG, model orchestration).
• Strong full-stack capability across apps, data stores, and infrastructure.
• Experience with AWS at scale and production systems.

Responsibilities

• Design and build AI-powered product features with end-to-end ownership.
• Own the full stack for deployed features including deployment, observability, cost, and security.
• Mentor and coach engineers on your team.
• Read existing systems to understand context before proposing changes.

Skills

Software engineering

AI systems

Full-stack

Mentoring

Production readiness

Education

Bachelor's Degree

Tools

AWS

Docker

Kubernetes

TypeScript

Python

Job description

The Role

We are looking for a pragmatic builder-architect — a senior engineer who ships fast without leaving a mess, and makes architectural choices that hold up as the product scales. This is a hands-on technical leadership role: roughly 40% writing code and prototyping, with the remainder spent on architecture, mentoring, and raising the engineering bar within your team.

You will embed with a product team for extended periods, owning technical direction and building AI-powered features end-to-end — from idea through production. You won’t just advise; you’ll build, and what you build will set the pattern for others. You may end up managing some engineers.

Most engineers lean one way. “Hackers” ship fast but accrue debt; “architects” build clean abstractions but stall on delivery. You are both. You know when to prototype loosely and when to invest in the durable version — and you can articulate why.

What You’ll Do

• Design and build AI-powered product features — agent architectures, RAG pipelines, model orchestration, evaluation frameworks, and guardrails — with the same engineering rigor as any production system: testable, observable, gracefully degrading.
• Own the full stack for the features you build — application code, data, infrastructure — making end-to-end decisions about deployment, observability, cost, and security.
• Make architectural choices that optimize for reversibility early and durability when the problem is actually understood.
• Mentor and coach engineers on your team, transferring judgment and mental models, not just answers. Calibrate involvement to stakes: get out of the way for cheap-to-reverse work, lean in for load-bearing decisions.
• Read existing systems as accumulated knowledge before treating them as debt. Understand why things are shaped the way they are before proposing changes.
• Identify and manage the blast radius of technical decisions — the dangerous ones at this level aren’t bad deployments, they’re bad directions.

What We’re Looking For

Engineering Judgment

• You think in failure modes and second-order effects, not happy paths and demos. “Who inherits this, and what does it cost them if I’m wrong?” is a question you ask naturally.
• You optimize for sustainability — testability, clear boundaries, sane defaults, documentation — so what you build can be owned and extended by others.
• You treat constraints as the design problem. You map what’s frozen, what’s validated, what other systems depend on, and what can’t take downtime before proposing solutions.

AI Engineering

• You have built and shipped AI-powered features in production — not just used AI tooling for personal productivity.
• You treat AI systems as engineering problems: versioned, evaluated, observable, and designed to degrade gracefully when models behave unexpectedly.
• You use AI as a force multiplier on judgment you already have — it accelerates the parts you understand well, precisely because you can evaluate the output.
• You use AI to compress the learning loop, not skip it. You build real mental models of new technology, using AI as an accelerant, not a crutch.

Working with Teams

• You transfer judgment, not just answers. You surface reasoning, install mental models, and make yourself progressively less necessary.
• You lead through demonstrated competence, not positional authority — and you know that doing the work yourself is sometimes the failure mode.
• You learn the team’s context, constraints, and history before injecting opinions. You earn trust by understanding what came before.

Learning

• You learn to a depth proportional to the decision. Evaluating something? Defensible opinion, move on. Committing the product to it? Deep enough to understand failure modes and sharp edges.
• When you pick up new technology, you’re trying to understand why it works the way it does and what problem its designers were solving — because that’s what transfers.

Technical Environment

Our stack spans the following. We don’t expect mastery of all of it — but given the role, you should be able to pick up almost any of it quickly.

• Cloud: AWS (primary) — architecture and infrastructure
• Front-end: React, TypeScript, Vite, Tailwind, shadcn/ui, BlockNote, Nginx
• Back-end: Node.js, Next.js, Python, Kafka, FastAPI, Dramatiq + Valkey (task queue), PynamoDB (single-table DynamoDB), SQLAlchemy async + asyncpg (PostgreSQL), Prometheus, MongoDB
• Infrastructure: Docker, Kubernetes (EKS), AWS CDK (TypeScript), ECS Fargate, DynamoDB, S3, RDS, Elasticache Valkey, Bedrock, Secrets Manager, SSM
• DevOps: NX, pnpm, GitHub Actions, GHCR, Docker/Buildx, Wiz, GitHub OIDC
• AI Infrastructure: Google ADK, LiteLLM, OpenRouter, Bedrock (KB, BDA, Guardrails), MCP, Langfuse + OpenTelemetry, RAGAS, specialist domain agents
• AI Tooling: Claude Code, Cursor, Copilot, opencode, Hermes Agent

Requirements

• Bachelor's Degree
• Minimum 8+ years of experience in software engineering, with meaningful time spent at a senior/staff/principal level owning technical direction.
• Production experience building AI-powered features — agent systems, RAG, model orchestration, or similar. Not just prompting or fine-tuning in isolation.
• Strong full-stack capability — comfortable across application code, data stores, and infrastructure. You don’t need to be an expert in all three, but you can’t treat any of them as someone else’s problem.
• Experience with AWS at scale — you’ve designed, deployed, and operated production systems on AWS, not just used it for personal projects.
• Demonstrated ability to mentor and elevate other engineers — through pairing, design review, or informal technical leadership.

Nice to Have

• Experience in regulated industries (healthcare, pharma, finance) where compliance constraints shape technical decisions.
• Background building multi-agent or agentic AI systems in production.
• Familiarity with infrastructure-as-code (CDK, Terraform, Pulumi) and container orchestration (EKS, ECS).
• Experience working in product-led engineering organizations where engineers own outcomes, not just outputs.

Are you ready to bring new insights and fresh thinking to the table? Fantastic! We have one seat available, and we hope it’s yours. Apply today.

Date Posted

04-Aug-2026

Closing Date

23-Aug-2026

Our mission is to build an inclusive environment where equal employment opportunities are available to all applicants and employees. In furtherance of that mission, we welcome and consider applications from all qualified candidates, regardless of their protected characteristics. If you have a disability or special need that requires accommodation, please complete the corresponding section in the application form.