Architecture
Architecture
Fund Analyst Intelligence is a production system for continuous fund validation.
It is organised as a pipeline with explicit workflow states.
It treats evidence and auditability as first-class concerns.
This page describes the conceptual architecture.
It focuses on components and responsibilities.
It is written to support implementation, integration, and operations.
Architectural goals
A production architecture must provide:
- repeatable monthly cycles with clear state transitions
- deterministic validation and delta computation
- evidence-first data structures and provenance links
- human review and approval controls
- reproducible reports and immutable cycle records
- observability and safe failure behaviour
High-level system view
A typical deployment contains the following subsystems:
- Ingestion and artefact store
- Extraction and normalisation
- Validation engine
- Snapshot and change log service
- Materiality and alerting
- Reporting and templating
- Review and approval workflow
- API layer and UI
- Observability and operations
Each subsystem has a single responsibility.
The boundaries exist to keep production behaviour predictable.
Core entities and flow
The pipeline is centred around three production records:
- Cycle: a monthly run with inputs, checks, decisions, and outputs
- Snapshot: the approved state of a fund at a point in time
- Artefact: a versioned source input used as evidence
A simplified flow is:
- cycle created
- artefacts ingested and registered
- extraction produces fields and claims linked to evidence
- validation produces checks and exceptions
- snapshot diff produces deltas
- materiality ranks changes and generates alerts
- reviewer resolves exceptions and approves updates
- reporting generates outputs from the approved cycle
- snapshot updated and cycle closed immutably
Subsystem details
1. Ingestion and artefact store
Responsibility
Capture inputs as versioned artefacts with metadata and access control.
Inputs
- documents, links, policy-approved sources
- portfolio membership and tagging (optional)
Outputs
- artefact inventory for the cycle
- provenance metadata and version relationships
Production requirements
- idempotent ingestion
- duplicate detection
- consistent naming and metadata validation
- retention and lifecycle controls
2. Extraction and normalisation
Responsibility
Convert artefacts into structured fields and claims linked to evidence.
Outputs
- extracted field set per fund
- claim list with evidence references
- confidence signals and extraction notes
Production requirements
- deterministic processing where possible
- stable field schemas and type constraints
- traceability from output back to source locations
3. Validation engine
Responsibility
Enforce completeness, consistency, freshness, and policy constraints.
Outputs
- validation results and check outcomes
- exception objects with severity and category
Production requirements
- versioned rule sets
- test coverage for core checks
- explicit failure modes and classification
4. Snapshot and change log service
Responsibility
Maintain approved fund snapshots and compute deltas.
Outputs
- structured delta sets vs prior snapshot
- change events with categories and evidence links
Production requirements - snapshots only on approval
- deterministic diff logic
- historical traceability and reproducibility
5. Materiality and alerting
Responsibility
Convert deltas and exceptions into prioritised queues and alerts.
Outputs
- ranked exceptions
- alert events with severity and payload standards
- follow-up breach escalations
Production requirements
- policy-defined thresholds
- low-noise alerting
- clear mapping to ownership and resolution workflow
6. Reporting and templating
Responsibility
Generate monthly memos, quarterly pack sections, and evidence packs.
Outputs
- reports with stable structure
- embedded provenance links
- approval stamps and cycle references
Production requirements
- template versioning
- reproducible generation from stored cycle record
- publication gated by approval state
7. Review and approval workflow
Responsibility
Support human-in-the-loop decisions with accountability.
Features
- explicit workflow states: draft → review → approved → published
- reviewer actions and rationale capture
- override tracking for key fields
Production requirements
- separation of duties where required
- immutable audit trail of decisions
- safe publication controls
8. API layer and UI
Responsibility
Expose stable contracts for the product UI and partner integrations.
Typical API surfaces
- fund profile and snapshot retrieval
- cycle status and artefact inventory
- exceptions and alerts
- report retrieval and export
- policy configuration (admin)
Production requirements
- contract versioning
- access control enforcement
- rate limits and predictable error codes
9. Observability and operations
Responsibility
Ensure the system is diagnosable and safe to operate.
Core practices
- structured logging with cycle ids and fund ids
- metrics: cycle durations, exception rates, failures
- error classification and runbooks
- controlled reruns with versioning
Production requirements
- safe failure without partial publication
- resumable cycles
- monitoring and alerting for job health
Deployment modes
Fund Analyst Intelligence can be deployed in stages:
- pilot mode with manual ingestion and standard outputs
- production mode with scheduled cycles and review workflow
- integrated mode with exports and APIs
- enterprise mode with deeper controls and operational SLAs
The architecture supports incremental adoption.
It is designed to avoid big-bang integrations.
Definition of done
The architecture is production-ready when:
- cycles are explicit and reproducible
- artefacts are versioned and policy-controlled
- validation and diffs are deterministic and tested
- reporting is gated by approval states
- audit trails capture decisions and evidence
- operations are observable with safe rerun behaviour
This is the standard Fund Analyst Intelligence is built to meet.