mini_researcher

Mini Researcher Example

This directory contains a complete example application: a mini research assistant that uses RAG (Retrieval-Augmented Generation) with validation.

Files

researcher.py

Main implementation of the research assistant with helper functions.

Key Features:

• Session management with caching
• Guardian model integration for safety
• Custom validation functions
• Word count constraints
• Subset validation for citations

context_docs.py

Document context and RAG document definitions.

init.py

Package initialization and exports.

Concepts Demonstrated

• Complete Application: Full-featured research assistant
• RAG Integration: Retrieval-augmented generation workflow
• Multi-Model Setup: Using different models for different tasks
• Validation Pipeline: Multiple validation requirements
• Safety Checks: Guardian model for content safety
• Custom Requirements: Domain-specific validation logic
• Caching: Efficient session reuse with @cache

Architecture

User Query
    ↓
Document Retrieval (RAG)
    ↓
Generation with Context
    ↓
Validation:
  - Word count check
  - Citation validation
  - Safety check (Guardian)
    ↓
Result

Key Components

Session Management

@cache
def get_session():
    """Get M session (change model here)."""
    return MelleaSession(backend=OllamaModelBackend(model_ids.IBM_GRANITE_4_MICRO_3B))

@cache
def get_guardian_session():
    """Get M session for the guardian model."""
    return MelleaSession(backend=OllamaModelBackend(model_ids.IBM_GRANITE_GUARDIAN_3_0_2B))

Custom Validation

def create_check_word_count(max_words: int) -> Callable[[str], bool]:
    """Generate a maximum-word-count validation function."""
    def cc(s: str):
        return len(s.split()) <= max_words
    return cc

def is_a_true_subset_of_b(a: list[str], b: list[str]) -> bool:
    """Check if a is true subset of b."""
    # Citation validation logic

Usage Pattern

This example demonstrates best practices for building production applications:

1. Separation of Concerns: Different modules for different responsibilities
2. Reusable Components: Helper functions that can be used across the app
3. Multi-Model Architecture: Using specialized models for specific tasks
4. Comprehensive Validation: Multiple layers of validation
5. Caching: Performance optimization with session caching

Related Documentation

• See rag/ for RAG examples
• See safety/ for Guardian examples
• See mellea/stdlib/requirements/ for validation patterns