Skip to content

36. Dual-Paradigm Framework (Symbolic vs Neural)

Mini-Project: Financial Compliance Checker

A dual-paradigm financial compliance system that routes open-ended policy questions to an LLM (System 1) and precise rule checks like CTR thresholds and wash-sale violations to a deterministic Python rule engine (System 2).

View on GitHub

Description

The Dual-Paradigm Framework explicitly separates an agent's reasoning into two paradigms and provides a meta-controller that decides which paradigm to use for each sub-task. The System 1 (neural/fast) path handles intuitive, pattern-matching tasks using the LLM directly. The System 2 (symbolic/slow) path handles deliberate, logical tasks using formal reasoning tools. This mirrors Kahneman's dual-process theory of human cognition.

Core Components

flowchart TD
    A[Input] --> B[Meta-Controller]
    B -->|Fast/Intuitive| C[System 1: Neural LLM]
    B -->|Slow/Deliberate| D[System 2: Symbolic Reasoning]
    C --> E[Output]
    D --> E

    D --> F[Logic Solver]
    D --> G[Constraint Checker]
    D --> H[Rule Engine]

    style B fill:#FF5722,color:#fff
    style C fill:#2196F3,color:#fff
    style D fill:#9C27B0,color:#fff

When to Use

  • Systems handling a mix of intuitive and analytical tasks
  • When you want to optimize compute by routing appropriately
  • Applications requiring both conversational AI and formal verification
  • When you want an explicit System 1/System 2 architecture