Agent Sandbox

This page demonstrates an experimental AI Agent interacting with a symbolic expert system implemented in Prolog.

This page is not a medical user interface; clinicians should instead use the Expert System, which is recommended to explore before using this sandbox.

Relationship between Agent and Expert System

Expert System

The Expert System is implemented entirely in Prolog and provides:

• a questionnaire interface for phenotype selection
• a knowledge base of rare diseases
• a phenotype database
• a symbolic reasoning engine implemented in Prolog

Disease descriptions can be explored here: https://diseases-non-interactive.amsafis.com/review

Agent Sandbox

The current page is only a sandbox demonstrating how an AI agent could interact with the expert system.

This sandbox does not contain the expert system.

Instead, it simulates how external software (for example a hospital application) might send a clinical note to an agent that interacts with the expert system.

Architecture demonstrated in this sandbox

Clinical note
(simulated hospital software)

        ↓

Agent (agent.amsafis.com)
phenotype normalization

        ↓

MCP request
task: diagnose

        ↓

Prolog reasoning engine
(diseases-non-interactive.amsafis.com)

        ↓

Candidate rare diseases

The phenotype normalization problem

Phenotype normalization is one of the major challenges in medical informatics.

Hospitals typically rely on standardized terminology systems such as:

• SNOMED CT
• Human Phenotype Ontology (HPO)
• ICD-10

Mapping clinical text to standardized phenotypes is a complex task and cannot be fully solved in a small demonstration system.

Machine-readable output

The agent returns a structured JSON output intended for software systems.

{
  "phenotypes": [
    "Myopia",
    "Hyperventilation"
  ],
  "candidate_diseases": [
    "biotinidase_deficiency",
    "fructose_1_6_bisphosphatase_deficiency",
    "holocarboxylase_synthetase_deficiency",
    "retinal_ciliopathy_due_to_mutation_in_usher_gene"
  ]
}

Test the agent

The form below simulates a clinical note sent by external software. For example, entering "Patient reports difficulty seeing distant objects and fast breathing" will generate the machine-readable output shown above.