Quiet Signals
Lab

Alexander Hepburn
Data Scientist & Scientific Programmer
Amsterdam, NL

I specialise in model diagnostics, applied machine learning theory, data visualisation and code-based typesetting. I study why models degrade in production, build RAG systems through rigorous retrieval optimisation, and develop publication-grade analytics using reproducible, code-based workflows.

Through my company, Quiet Signals Lab, I design transparent and robust tech products for both business and research contexts.

Services

Your model degrades in production, but why? I systematically audit your model and data to identify failure modes—diagnosing distribution shifts, data quality issues, and what patterns your model actually learned. Through controlled ablation studies, I isolate which factors matter most.

Deliverables:

  • Root cause diagnosis with quantified impact analysis
  • Evidence-based recommendations, ranked by priority
  • Ablation studies proving which fixes actually work
  • Documentation of failure modes and monitoring thresholds

I design and build automated reporting pipelines using Quarto that transform your data into publication-quality reports and dashboards. Reports are reproducible, version-controlled, and easy to maintain—no locked files, no manual updates. This scales analytics without hiring a full-time team.

Deliverables:

  • Reproducible reports in any format (HTML, PDF, interactive)
  • High-quality graphics and professional typesetting
  • Open-source, code-driven workflows (no proprietary software)
  • Efficient Python-powered data transformation and formatting

I build RAG systems with proven strategies to minimise hallucination. This means getting retrieval right—sentence-aware chunking, semantic reranking, strict generation constraints—so every answer cites its sources. Backend uses Python (FastAPI) and local or cloud vector storage (ChromaDB), supporting both local models (Ollama) and commercial providers (OpenAI, Mistral).

Deliverables:

  • Production-ready RAG backend with REST API
  • Vector database setup and document ingestion
  • Retrieval optimisation (chunking, embedding tuning, reranking)
  • Hallucination safeguards with source attribution
  • Documentation and handover for operation

Tech & Methods

Python PyTorch PySpark PostgreSQL AWS ChromaDB Plotly Causal Inference A/B Testing Research Methodology Machine Learning Theory Privacy-First Architecture

I avoid one-size-fits-all approaches. Every project gets the balance between state-of-the-art techniques and practical, maintainable solutions that fit your constraints.

Get in Touch View on GitHub