Senior AI Systems Quality Engineer

Abacus Insights•Boston, MA

2h•Remote

About The Position

At Abacus Insights, we’re building AI‑powered platforms that sit at the center of critical healthcare decisions—where reliability, correctness, and trust matter as much as innovation. As our use of agentic, LLM‑driven systems scales, ensuring these systems behave safely and predictably in real‑world environments isn’t optional—it’s foundational. The Senior AI Systems Quality Engineer plays a critical role in making our AI systems production‑ready, operable, and trustworthy. This is not a traditional QA role. Instead, you’ll help define how quality is engineered into agentic systems from the very beginning—embedding guardrails, evaluation signals, and safe‑failure behavior directly into how these systems are designed, built, and deployed. In this role, you will also help design and own an AI testing platform that operates natively within our Databricks Medallion architecture—leveraging tools like MLflow to enable evaluation, lineage, traceability, and auditability at scale. This is a platform-oriented role, focused on establishing repeatable, scalable quality practices rather than point-in-time validation. You’ll partner closely with AI Engineers, platform teams, and delivery stakeholders to design custom evaluation frameworks, validation pipelines, and automated testing harnesses that reflect how agentic systems behave in production. You’ll focus on validating orchestration logic, managing non‑deterministic behavior, and ensuring AI systems degrade safely under uncertainty, scale, and failure. This role is automation-first by design. You will build end-to-end automated AI testing integrated into the development lifecycle, where validation is continuously enforced—not manually reviewed after the fact. Your work will directly influence release readiness, operational confidence, and Abacus’ ability to responsibly deploy advanced AI in mission‑critical healthcare environments. If you believe quality is a core engineering discipline—not a downstream checkpoint—this role gives you the opportunity to shape how reliable, governed, and scalable agentic AI is built from the ground up.

Requirements

7+ years of software engineering experience, primarily in backend or platform systems.
Proven experience designing and implementing AI testing automation in production environments, not just executing tests.
Demonstrated ability to build custom validation, evaluation, or testing frameworks for complex, distributed systems.
Strong proficiency in Python and/or TypeScript within modern AI engineering stacks.
Hands-on experience with AI-powered systems, including LLM-based or agentic workflows and non-deterministic behavior.
Experience designing or contributing to AI testing at scale, including regression frameworks, long-tail evaluation, and large test coverage.
Deep understanding of CI/CD integration, including embedding automated tests and quality gates into deployment pipelines.
Solid understanding of AWS cloud-native architectures.
Track record of engineering for quality, reliability, governance, and safety as core system design principles.
Working knowledge of security, privacy, and operational risk in regulated or mission-critical environments, including failure modes and recovery.
Experience with AI testing methodologies, including evaluation of non-deterministic outputs, drift detection, bias/fairness testing, and robust regression strategies.
Proven ability to establish measurable trust thresholds for AI systems, including defining and operationalizing success metrics such as query accuracy, hallucination limits, explainability, and PHI-safe behavior as enforceable release criteria.
Experience working with domain experts to define correctness and real-world validation scenarios, enabling large-scale, business-relevant test coverage that reflects true production use cases rather than engineering-only perspectives

Nice To Haves

Experience with Databricks-native environments and Medallion architecture.
Experience using MLflow for model evaluation, lineage tracking, and auditability.
Exposure to observability tools (e.g., Datadog, Prometheus, Grafana) for distributed AI workflows.
Familiarity with LLM evaluation techniques, guardrails, and policy enforcement frameworks.
Experience evaluating performance, latency, or cost regressions in AI systems.
Ability to clearly document system behavior and quality trade-offs for technical and business audiences.
Formal training or certification in AI/ML systems (e.g., ISTQB AI Testing, AWS ML Specialty, Google ML Engineer).
Experience designing and iterating on prompts, agent behaviors, and orchestration logic as versioned, testable artifacts, enabling rapid refinement of AI system behavior (often referred to as “vibe coding”) without heavy reliance on traditional code changes.
Familiarity with using AI systems to generate and expand test scenarios, including creating large-scale, diverse, and adversarial test datasets to significantly improve coverage and validation depth.

Responsibilities

Build and ship production-grade, automated validation frameworks, test harnesses, and evaluation pipelines across the AI lifecycle (design → deploy).
Design and evolve an AI testing platform integrated with Databricks and MLflow, enabling repeatable testing, traceability, and auditability.
Create large-scale, scenario-based test suites (hundreds to thousands of cases) to validate agentic workflows end-to-end, including edge cases, long-tail scenarios, and failure modes.
Validate orchestration behavior (tool use, memory, decision logic) and stress-test non-deterministic system behavior before production.
Embed quality by design: define system contracts, guardrails, and safe-degradation patterns at key boundaries.
Define measurable quality signals for LLM systems (grounding, hallucinations, relevance, latency, cost) and integrate them into CI/CD pipelines as automated quality gates.
Ensure AI validation runs automatically on model, prompt, and code changes—enabling continuous quality enforcement.
Build reusable libraries and components so teams can adopt consistent AI quality practices quickly.
Own aspects of AI release readiness, including defining go/no-go criteria based on measurable quality thresholds.
Partner with AI, platform, security, and delivery teams to translate mission needs into clear quality criteria, tradeoffs, and confidence levels.