Architecture

A production architecture for Fund Analyst Intelligence: ingestion, extraction, validation, change detection, reporting, and governance.

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:

  1. Ingestion and artefact store
  2. Extraction and normalisation
  3. Validation engine
  4. Snapshot and change log service
  5. Materiality and alerting
  6. Reporting and templating
  7. Review and approval workflow
  8. API layer and UI
  9. 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:

  1. cycle created
  2. artefacts ingested and registered
  3. extraction produces fields and claims linked to evidence
  4. validation produces checks and exceptions
  5. snapshot diff produces deltas
  6. materiality ranks changes and generates alerts
  7. reviewer resolves exceptions and approves updates
  8. reporting generates outputs from the approved cycle
  9. 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.