Feature: SMS-to-Transaction Mapping for Smart Transaction Templates

[RadCod3]

Feature Request: SMS-to-Transaction Mapping for Smart Transaction Templates

Problem Statement

When logging expenses by copying and pasting bank SMS messages, users must manually specify transaction details every time. This becomes repetitive for similar transactions from the same merchants or transaction patterns, leading to:

• Time-consuming manual entry for recurring transactions
• Inconsistent categorization of similar expenses
• No learning from previous transaction patterns

Proposed Solution

Implement a fuzzy-searchable index that maps SMS content to transaction templates, enabling the system to suggest transaction details based on historical patterns.

Core Requirements

Fuzzy Search Capability

• Match SMS patterns even when exact text varies (amounts, dates, reference numbers)
• Identify similar merchants despite minor variations in naming
• Handle SMS format differences across banks or time periods
• Score similarity confidence for multiple potential matches

Learning and Storage

• Automatically store SMS → Transaction relationship after each transaction creation
• Build a searchable index of historical SMS patterns and their corresponding transactions
• Learn from user preferences and corrections over time
• Maintain transaction templates that can be suggested for similar SMS content

Smart Suggestions

• When user provides SMS text, query the index for similar historical patterns
• Present suggested transaction templates with confidence scores
• Allow users to accept, modify, or reject suggestions
• Use suggestions as defaults while preserving manual override capability

User Workflow Enhancement

Current Workflow

1. User copies bank SMS
2. User manually enters all transaction details (account, category, budget, etc.)
3. Transaction created

Enhanced Workflow

1. User copies bank SMS
2. System analyzes SMS and finds similar historical patterns
3. System suggests transaction template(s) based on matches
4. User reviews suggestions and accepts/modifies as needed
5. Transaction created and SMS pattern stored for future use

Example Scenarios

Scenario 1: Recurring Merchant

• First SMS: "HDFC Bank: Rs.250 debited from A/c XX1234 for STARBUCKS COFFEE on 15-Jan-24"
• System: No matches found, user creates transaction manually
• Future SMS: "HDFC Bank: Rs.320 debited from A/c XX1234 for STARBUCKS COFFEE on 22-Jan-24"
• System: Suggests similar transaction template (same merchant, expense category, possibly same budget allocation)

Scenario 2: Similar Transaction Types

• Pattern: Multiple grocery store transactions with different amounts and stores
• System: Learns that grocery-related SMS typically map to "Groceries" category and "Food" budget
• Suggestion: Auto-suggest appropriate category and budget for new grocery transactions

Scenario 3: Utility Bills

• Pattern: Monthly electricity bill SMS with varying amounts
• System: Recognizes utility payment pattern and suggests "Utilities" category and "Bills" budget
• Benefit: Consistent categorization despite amount variations

Success Criteria

Accuracy

• Suggest relevant transaction templates for at least 70% of SMS inputs after initial learning period
• Maintain high precision to avoid incorrect suggestions that confuse users
• Improve suggestion quality over time with more training data

User Experience

• Reduce manual entry time for repeated transaction types
• Maintain transaction creation speed even when no suggestions available
• Allow easy acceptance or rejection of suggestions
• Provide clear confidence indicators for suggestions

System Performance

• Fast fuzzy search responses (< 500ms for suggestion retrieval)
• Efficient storage that scales with transaction volume
• Minimal impact on existing transaction creation workflow

Possible Solutions

Text Similarity Approaches

• N-gram matching: Compare SMS text patterns using character/word n-grams
• Semantic similarity: Use embeddings to find conceptually similar SMS content
• Merchant extraction: Focus matching on extracted merchant names and transaction types
• Hybrid approach: Combine multiple similarity metrics for better accuracy

Storage and Indexing

• Full-text search: Enable fuzzy text search across historical SMS content
• Merchant normalization: Store normalized merchant names for better matching
• Pattern templates: Extract and store common SMS patterns rather than exact text
• Weighted scoring: Consider recency, frequency, and user feedback in suggestions

Integration Points

• Enhance existing transaction creation MCP tools to accept SMS input
• Add new MCP tool specifically for SMS-based transaction creation
• Integrate with current budget allocation and categorization logic
• Maintain backward compatibility with existing workflows

Implementation Considerations

Privacy and Data

• Store SMS content locally within LamPyrid instance
• Ensure sensitive information (account numbers, exact amounts) handled appropriately
• Allow users to delete or modify stored patterns
• Consider data retention policies for old mappings

Configuration

• Allow users to enable/disable SMS suggestion feature
• Configure similarity thresholds for suggestions
• Set maximum number of suggestions to present
• Option to auto-apply high-confidence suggestions

Maintenance

• Provide tools to view and manage stored SMS patterns
• Allow manual correction of mappings
• Handle cleanup of outdated or incorrect patterns
• Export/import capabilities for backing up learned patterns

Expected Benefits

• Faster expense logging: Reduce time spent on repetitive transaction entry
• Improved consistency: Better categorization and budget allocation for similar transactions
• Enhanced user experience: More intelligent and adaptive personal finance management
• Learning system: Becomes more useful over time as it learns user preferences

This feature would significantly improve the primary use case of LamPyrid - efficiently logging expenses from bank SMS notifications while maintaining accuracy and consistency.