Alex Baas Principal AI Engineer · Production LLM Agents

Health and longevity company (clinically researched supplements, DNA and biological-age testing) handling regulated consumer-health and genetic data.

• Led design and prototyping of an agentic DNA-insights assistant over customers' DNA test results: multi-agent orchestration with tool calling (MCP), personalised genomic insights and interactive visualisation.
• Built GraphRAG over a knowledge graph of genomics research (PubMed papers to genes, variants, nutrients, outcomes): hybrid vector + graph retrieval, reranking and source citations grounding every answer in peer-reviewed evidence.
• Built a recommendation agent matching DNA results to best-suited supplement formulations, powering personalisation and the sales funnel.
• Engineered LLM guardrails for regulated health content: health-claim compliance (FDA/FTC structure-function rules, EU Nutrition and Health Claims Regulation), medical-advice refusal, prompt-injection defense and output verification.
• Implemented privacy-by-design for genetic data (GDPR special category, HIPAA-aligned): PII/PHI redaction before LLM calls, consent-scoped retrieval and audit trails for AI governance.
• Set up LLM evaluation and observability as release gates: golden datasets, LLM-as-a-judge grounding and faithfulness scoring, end-to-end tracing.
• Owned end-to-end delivery: designed, built and deployed AI services on AWS and GCP with Docker, Kubernetes, Terraform, Ansible and CI/CD.
• Built the ML-ready data foundation (BigQuery, dbt) and designed churn-prediction models (XGBoost, LightGBM) feeding personalisation.

eQMS SaaS for life sciences (quality management for pharma, biotech and medical-device companies) in a GxP-regulated environment. Fixed-scope engagement: design and build agentic AI into the product.

• Led engineering of the LLM platform service (Python, FastAPI, AWS Bedrock / Claude): REST APIs mirrored as MCP tools, OpenTelemetry LLM observability, per-call token and cost budgets, signed and versioned prompts with per-response provenance.
• Designed the product's agentic AI chat: one agent with a growing set of skills over QMS data, permission-aware answers with citations, human-in-the-loop by design and a full audit trail. Built its first release.
• Built eval-driven release gates in CI/CD: SME golden datasets, rubric-based LLM-as-a-judge juries calibrated against human reviewers, per-metric floors. Shipped grounded document-AI extraction and generative assessment features behind them.
• Architected the end-to-end RAG and agent stack: document ingestion (parsing, OCR, chunking, embeddings), context engineering, governed tool calling and output verification; designed multi-tenant, permission-aware hybrid RAG with citations.
• Conducted LLM security red teaming (OWASP LLM Top 10, STRIDE; Promptfoo, Garak, PyRIT) and designed the mitigations and guardrails.
• Designed the AI quality and governance framework: evals, observability and compliance evidence mapped to the EU AI Act, ISO/IEC 42001, GAMP 5 and FDA 21 CFR Part 11.

AI-driven platform for digital marketing campaigns.

• Led ML engineering for campaign intelligence: predictive models (PyTorch, scikit-learn) for audience targeting and campaign-performance forecasting, deployed to production.
• Prototyped generative AI for marketing content on GPT-3 and early ChatGPT APIs (2022): prompt engineering and LLM content workflows.
• Architected the PostgreSQL data layer and pipelines feeding model training and analytics; refactored Python and Go services for a 50% server performance gain.

Digital coaching platform for enterprise employees.

• Built NLP-powered coach matching and content recommendation (TensorFlow), personalising the coaching experience across the platform.
• Productionised ML features in Python and Node.js microservices on AWS and GCP with Docker, Kubernetes and CI/CD.
• Architected the REST and GraphQL APIs connecting ML services to the product.

Leading European healthcare booking and telehealth platform.

• Developed NLP models (TensorFlow, PyTorch) for patient-doctor message classification and routing, working with sensitive health data under GDPR.
• Built end-to-end product features for healthcare operations across Ruby on Rails, Go and React.