AI Portfolio Assistant System Architecture
Diagram
Key
User Layer
| Component | Purpose |
|---|
| Portfolio Manager | Human participant in the governed workflow. Retains authority for activities requiring human judgement or approval. |
FDC3 Applications
| Component | Purpose |
|---|
| AI Portfolio Assistant | The primary subject of governance in the demo. Consumes FDC3 context and raises FDC3 intents to use capabilities supplied by other applications. |
| Portfolio App | Displays portfolios, positions and holdings, and shares the user's current portfolio selection through FDC3. |
| Risk App | Provides deterministic portfolio risk analysis, including exposure, concentration and stress scenarios. |
| Research App | Provides research and supporting information about portfolios and instruments. |
| Order Entry / Trade Blotter | Receives proposed orders for human review and, where authorised, submission. |
| Governance Console | Allows users to inspect governed activities, decisions and supporting evidence. |
Desktop Agent
| Component | Purpose |
|---|
| Context Routing | Routes FDC3 contexts such as the selected portfolio or instrument between applications. |
| Intent Resolution | Resolves FDC3 requests for capabilities to applications capable of handling them. |
| Governance Enforcement Point | Ensures sensitive requests are checked by the Governance Gateway before the Desktop Agent forwards them. |
Governance Runtime
| Component | Purpose |
|---|
| Governance Gateway | Evaluates proposed sensitive activities and returns decisions such as ALLOW, DENY or REQUIRE APPROVAL. |
| Deterministic Policy Evaluation | Evaluates objective rules such as permissions, approved models, data freshness and explicit authority limits. |
| Semantic / AI Evaluation | Evaluates requirements requiring interpretation or judgement. |
| Human Approval Workflow | Routes activities requiring explicit human authority or approval. |
| Evidence Recording | Writes evaluations, decisions and supporting evidence to the Decisions and Evidence Record (object store). |
Governance Model
| Component | Purpose |
|---|
| Gemara Artifacts | Machine-readable artifacts describing governance requirements and their relationships. |
| AI Management System (AIMS) | The organisational governance framework within which the AI is managed, aligned to ISO/IEC 42001. |
| Sensitive Activities | Activities subject to governance, such as recommendation generation, order construction and trade execution. |
Financial Data and Services
| Component | Purpose |
|---|
| Portfolio / Position Data | Source data containing portfolio holdings and positions. |
| Market Data | Pricing and other market information used for analysis and risk calculations. |
| Research / News / Filings | Information sources used by the Research App and AI Assistant. |
| Trading / Execution Service | External capability through which authorised orders can ultimately be submitted or executed. |
Architectural Separation
The important architectural boundary is:
FDC3 applications make ordinary FDC3 requests; the Desktop Agent enforces governance; the Governance Gateway makes the decision.
That keeps the AI Assistant and the individual business applications free from embedded governance logic, while ensuring consequential inter-application actions cannot bypass the governance layer.