Data, Models and Reasoning for High-Stakes Decisions

amsafis is a specialised consultancy focused on transforming complex data into reliable predictions, interpretable logic and actionable knowledge.

The company is small by design: the work is carried out directly by a senior expert with more than three decades of professional experience in modelling, statistics, operational analytics and intelligent systems.

This ensures rigor, personal responsibility and solutions that can be understood, validated and trusted.

Three complementary pillars

Modern decision support requires more than one technique.
At amsafis, all projects combine one or more of these pillars, depending on the problem.

1. Machine Learning — Predictive Modelling and Pattern Discovery

Machine Learning provides the mathematical backbone for prediction, risk scoring, anomaly detection and non-linear pattern discovery.

It is particularly effective when real-world data include noise, interaction, heterogeneity or limited sample sizes.

Examples of techniques:

ML is used both for direct prediction and as a source of quantitative evidence that later feeds Expert Systems or rule-based reasoning.

➡️ Example: Association rules identifying genetic patterns linked to penicillin allergy.

2. Expert Systems — Transparent Rules and Explainable Logic

Expert Systems convert clinical, industrial or organisational knowledge into explicit rules that can be audited, traced and verified.

This approach provides clarity and full explainability, which is essential in regulated environments such as healthcare, manufacturing quality or operational safety.

These systems support:

An example is the rare disease sandbox, which uses knowledge rules to identify possible diagnoses in interactive mode. Explore the interactive demonstration.

The same knowledge engine can be accessed both in a human-driven way and integrated into automated processes that allow the knowledge to be reused as an executable service within decision architectures or computational assistance.

3. RAG-LLM — Retrieval-Augmented Understanding of Documents

Large Language Models are only reliable when connected to verified sources.

RAG pipelines link LLMs with scientific articles, guidelines, technical documents or internal reports, ensuring:

This makes it possible to query hundreds of pages of technical or clinical documentation in natural language, with transparent references and controlled behaviour.

RAG Sandbox using ChatGPT-3.5

Integrated decision support

We integrate Machine Learning, Expert Systems, and RAG-LLM to provide robust and interpretable decision support across domains.

These components can be combined into task-oriented Agents that coordinate reasoning, prediction, and contextual understanding within a single system.

This integrated approach enables flexible, traceable, and domain-adapted solutions for complex, high-stakes environments. Learn more about our Agent approach.

Areas of work

How we work

Projects are carried out with direct technical involvement at every stage:
problem definition, modelling, validation, implementation and integration.

The priority is always the same:
precision, interpretability and systems that organisations can trust.

If you want to explore a project

You can get in touch to discuss your needs, constraints or data characteristics.

The work can involve Machine Learning, Expert Systems, RAG-LLM or a combination of the three, depending on the context.

A structured explanatory wiki describing the three pillars and their interaction is available here.

Consultant Wiki.