Architecture Overview¶

FIML (Financial Intelligence Meta-Layer) is built with a modular, scalable architecture designed for high performance and extensibility.

System Layers¶

Client Layer¶

Supports multiple client types: - ChatGPT plugins - Claude Desktop integration - Custom applications - Telegram bots

API Gateway¶

Unified MCP API with: - Request routing - Authentication - Rate limiting - Compliance checks

Data Arbitration Engine¶

Intelligent provider selection based on: - Availability (30%) - Freshness (25%) - Reliability (25%) - Latency (15%) - Cost (5%)

Multi-Agent Framework¶

Ray-based distributed agents: - Fundamentals analysis - Technical analysis - Macro analysis - Sentiment analysis - News aggregation

Provider Layer¶

Multiple data sources: - Yahoo Finance - Alpha Vantage - Financial Modeling Prep (FMP) - CCXT (crypto exchanges) - Custom providers

Project Structure¶

FIML/
├── fiml/
│   ├── server.py                  # Main FastAPI server
│   ├── arbitration/               # Data Arbitration Engine
│   ├── core/                      # Core utilities (Config, Logging, Models)
│   ├── mcp/                       # MCP Protocol Implementation
│   ├── providers/                 # Data Provider Abstraction
│   ├── cache/                     # Cache Layer (Redis + Postgres)
│   ├── dsl/                       # FK-DSL Parser & Executor
│   ├── agents/                    # Multi-Agent Orchestration
│   └── compliance/                # Compliance Framework
├── config/                        # Configuration files (Prometheus, Grafana)
├── k8s/                           # Kubernetes manifests
├── scripts/                       # Helper scripts
└── tests/                         # Test suite

Module Responsibilities¶

Core (`fiml/core/`)¶

Configuration management
Domain models and types
Logging infrastructure
Custom exceptions

MCP (`fiml/mcp/`)¶

FastAPI routing for MCP protocol
Tool implementations (search-by-symbol, search-by-coin, etc.)
Request/response handling

Providers (`fiml/providers/`)¶

Abstract base provider interface
Provider registry and lifecycle management
Individual provider implementations
Health monitoring

Arbitration (`fiml/arbitration/`)¶

Provider scoring algorithm
Execution plan generation
Auto-fallback logic
Multi-provider data merging

Cache (`fiml/cache/`)¶

L1 (Redis) in-memory cache
L2 (PostgreSQL/TimescaleDB) persistent cache
Predictive cache pre-warming

Agents (`fiml/agents/`)¶

Ray-based distributed orchestration
Specialized worker agents
Task routing and coordination

Key Design Principles¶

Provider Agnostic: Easy to add new data sources
Cache-First: Multi-tier caching for performance
Fault Tolerant: Automatic fallback and retry
Observable: Comprehensive monitoring and logging
Compliant: Built-in regional compliance

Security¶

Security features include: - API key management and rotation - Rate limiting and throttling - Request authentication - Compliance checks per region - Audit logging

For more details on security and compliance, see the Compliance Framework.