Symbolic knowledge architecture
At the core of sAÏmone sits a structured knowledge base encoding pharmaceutical Medical Affairs domain logic: strategic pillars, tactical taxonomies, stakeholder profiles, evidence hierarchies, and metric definitions — all with explicit interdependencies, constraint rules, and feasibility gates. This is not learned from training data. It is authored, validated, and deterministic.
Constraint-flow validation
Every output passes through a multi-step constraint chain. Stakeholder access levels gate which tactics are eligible. Tactics must align with their parent strategic pillars. Claims are capped by evidence source tiers — a regulatory approval claim requires tier-one regulatory sources, not inferred data. If any link in the chain breaks, the output is blocked. This is how hallucination is structurally prevented, not just discouraged.
Evidence epistemology
Every data point carries an explicit verification status: verified against primary sources, inferred from validated data, or flagged for internal confirmation. A four-tier evidence hierarchy governs what each source is allowed to claim — peer-reviewed publications support outcome-level claims, while preprints are limited to activity-level observations. Cross-validation rules require multi-source confirmation for critical assertions.
Computational strategy layer
Beyond qualitative frameworks, sAÏmone runs live quantitative analysis within sessions: Monte Carlo simulations for regulatory timeline uncertainty, game-theoretic modelling for competitive scenarios, decision-theoretic reasoning under incomplete information, and mechanism design for incentive structuring. These are not summaries of theory — they are executable computational models producing probability-weighted recommendations.
How It Works
The Symbolic Layer
The neural model handles language — understanding your brief, reasoning across evidence, generating prose. The symbolic layer handles structure — defining what entities exist, what relationships are valid, what claims each source can support, and what constraints must hold. The neural model cannot override the symbolic layer. It operates within it.
Deterministic
Auditable
Constraint-Gated
Medical Affairs Intelligence Platform
