AI Native Engineer

About Unilabs:

Headquartered in Geneva and part of the A.P. Møller Group, Unilabs is one of Europe’s leading medical diagnostics companies, offering a complete range of laboratory, pathology, genetics, and imaging services to patients across 14 countries. Unilabs invests heavily in technology, equipment, and people – using digital technologies in its state-of-the-art laboratories and imaging institutes – to improve the lives of close to 100 million people every year.

About the job:

We're looking for an AI Native Engineer who wants to solve complex, high-impact challenges with LLMs, agentic frameworks, and modern AI tooling. In this role, you'll design and deploy production-grade AI systems that transform millions of unstructured pathology and genomics records into actionable clinical insights, helping shape the future of precision medicine. Join a leading European diagnostics organization where your work will directly influence healthcare innovation and patient outcomes at scale.

Core Responsibilities

1. Core Agentic Architecture & Retrospective Extraction:

• LLM Extraction Agents: Design, build, and maintain production-grade LLM-based extraction pipelines to automatically parse years of unstructured PDF pathology reports.
• Structured Parsing: Programmatically extract clinical entities such as diagnoses, tumor grades, pathological staging, and critical biomarker statuses from raw, free-text documents.
• Framework Selection: Evaluate and integrate specialized agentic frameworks and orchestration tooling (e.g., LangChain, LlamaIndex, or direct LLM API implementations) based on measurable extraction accuracy against real-world clinical text, rather than what is fashionable.
• Confidence Scoring & Human-Review Loops: Build programmatic confidence scoring systems and human-inthe- loop validation queues that flag low-confidence extractions for clinical review based on validation parameters defined by our Clinical Informatics Lead.

2. Multi-Modal Pipeline & Next-Gen API Infrastructure:

• Diagnostic Data Fusion: Architect and maintain the data pipelines that link pathology LIS data with separate molecular/genetics information systems. You will ensure that vital markers like KRAS, NRAS, BRAF, MMR/MSI status, and ctDNA results seamlessly map to the exact same case record as the histology diagnosis.
• Interoperable Interface Engineering: Implement robust REST APIs, HL7 v2, or HL7 FHIR interfaces to feed structured pipelines directly into downstream matching layers or ecosystems like Proscia Concentriq and Aperture.
• Future Ecosystem APIs: Lay the architectural groundwork for secure, high-throughput API layers destined to interface with premium consumer wearables, external preventive health apps, and cloud-native hospital systems.
• Data Quality Observability: Develop automated data-quality monitoring systems to catch and flag anomalous outputs, missing biomarker fields, or incomplete clinical records before they touch delivery endpoints.

3. Governance, De-Identification & Compliance:

• Anonymization Infrastructure: Implement technical de-identification protocols to securely strip or pseudonymize direct and indirect patient identifiers.
• Regulatory Alignment: Technical execution must align completely with strict health data privacy guardrails across global and regional frameworks, including the Swiss nDSG and EU GDPR Article 9.
• Lineage Tracking: Build exhaustive audit logging and data lineage tracking for every clinical record processed, preserving clinical data provenance for pharma and clinical partner credibility.

Requirements

AI Native & Agentic Mindset

• LLM Engineering Pro: Practical, hands-on experience utilizing LLM APIs, building system prompt state machines, and fine-tuning prompt engineering for highly structured text-extraction tasks.
• Agent Infrastructure Fluency: Direct experience working with agentic frameworks (LangChain, LlamaIndex, or equivalent custom graph state setups) to orchestrate complex, multi-step clinical data transformation workflows.
• Production Focus: You have shipped non-deterministic models into production environments and understand how to manage context windows, token costs, rate limits, and output evaluation metrics.

Core Software Engineering & Stack Experience

• Backend Proficiency: 4–7+ years of core software engineering experience with deep mastery of Python and SQL, capable of debugging asynchronous, multi-step pipelines independently.
• Regulated API Design: Deep familiarity with constructing and consuming production-grade REST APIs within highly regulated or clinical environments.
• Cloud & Containerization: Practical deployment experience across cloud infrastructure providers (AWS, Azure, or GCP) utilizing Docker containerization.
• Data Standards (Highly Preferred): Working knowledge of clinical health standards like HL7 v2, FHIR, or relational data models such as OMOP CDM and CDISC conventions.
• Data Formats (A Plus): Exposure to digital pathology data formats (DICOM, whole slide image file formats like SVS and NDPI), or LIS systems.

Benefits

What We Offer

• Hybrid working model ( office & remote flexibility)
• International, collaborative, and regulated product environment
• Competitive compensation and benefits
• Long-term ownership of a strategic healthcare product
• The Ultimate Unfair Data Moat: Direct engineering access to Europe's largest diagnostic pool—combining deep Pathology, Imaging, and Blood tests across millions of real, longitudinal patient journeys.
• No Toy Problems: The opportunity to move past generic chatbot wrappers and deploy agentic AI that directly impacts precision clinical trial execution, therapeutic drug development, and global preventative longevity markets.
• True Entrepreneurial Ownership: The execution speed, raw ownership, and equity upside of a venturebacked standalone seed-stage company, powered by the structural footprint of Unilabs and A.P. Møller Holding.

Working Environment Expectation

AI-Assisted Workflow: We build with modern tooling. You are expected to comfortably utilize AI-assisted environments like Cursor, GitHub Copilot, or equivalent editors as an active force multiplier to accelerate problem-solving. We care about what you ship, not how many characters you manually type.