System Architecture

Six explicit layers separate presentation, data, reasoning, governance, and evaluation. Each layer can be inspected, audited, and replaced independently.

Frontend

— React + TanStack Start

Layer 1/6

Clinical intake form (TH/EN)
AI Analysis & Evidence Panel
Clinician review & override UI
Case report export (Markdown / PDF)

Backend

— Lovable Cloud (managed Postgres + Auth)

Layer 2/6

patient_cases, ai_results
clinician_observations
case_outcomes
expert_validations

Knowledge Base

— Curated TTM ontology tables

Layer 3/6

explainability_traces
therapeutic_directions
clinical_safety_rules
knowledge_consensus
ml_feature_schema / ai_response_schema

AI / Reasoning Layer

— Rule-based syndrome prediction

Layer 4/6

Deterministic rules over symptom features
Outputs syndrome_id, risk_level, referral
Confidence + reasoning text
Hook-points for future ML models

Governance Layer

— Safety + clinician oversight

Layer 5/6

Safety rule triggers → urgent_referral
Clinician override always wins
Override comparison & audit log
Mandatory disclaimers

Validation Layer

— Expert evaluation

Layer 6/6

1–5 ratings across 5 dimensions
Reviewer role + free-text comments
Dashboard aggregates by dimension/role
Drives thesis evaluation

Data flow


  Clinician input  ──▶  Frontend form
        │
        ▼
  patient_cases  ──▶  Rule engine  ──▶  ai_results (syndrome_id, risk, confidence)
                                          │
                                          ▼
                          Knowledge base lookup (syndrome_id)
                            ├─ explainability_traces
                            ├─ therapeutic_directions
                            ├─ clinical_safety_rules  ──▶  Safety escalation?
                            └─ knowledge_consensus
                                          │
                                          ▼
                              Evidence Panel (UI)
                                          │
                                          ▼
                       Clinician review / override  ──▶  clinician_observations
                                          │
                                          ▼
                         Case report (Markdown / PDF)
                                          │
                                          ▼
                       Expert validation  ──▶  expert_validations