How Intelligence Works¶

AI-Augmented Email Analysis for TAMs

Overview¶

TAMINATOR uses pattern-based AI analysis to extract structured information from unstructured emails. Unlike external AI services, TAMINATOR runs 100% offline using local pattern matching and keyword analysis.

Philosophy: Hybrid Intelligence¶

Taminator combines multiple intelligence approaches:

1. Pattern Matching (Primary)¶

• Regex patterns for case numbers, emails, phones
• Keyword detection for issue types
• Domain-to-customer mapping
• Proven, deterministic, fast

2. Context Analysis (Secondary)¶

• Email signature parsing
• Organizational hierarchy detection
• Urgency indicator detection
• Deadline extraction

3. Historical Learning (Future)¶

• Feedback from TAM corrections
• Accuracy improvement over time
• Custom pattern development
• Team intelligence sharing

Analysis Pipeline¶

graph LR
    A[Email Text] --> B[Pattern Extraction]
    B --> C[Case Number]
    B --> D[Customer ID]
    B --> E[Contacts]
    B --> F[Issue Type]
    B --> G[Urgency]
    C --> H[Intelligence Report]
    D --> H
    E --> H
    F --> H
    G --> H
    H --> I[Recommended Actions]

Step 1: Pattern Extraction¶

Case Number Detection:

patterns = [
    r'case[#\s]*(\d{8})',
    r'case\s+number[:\s]*(\d{8})',
    r'\b(\d{8})\b',  # 8-digit number
]

Customer Identification:

# Domain mapping
'@jpmchase.com' → 'JP Morgan Chase'
'@wellsfargo.com' → 'Wells Fargo'
'@bankofamerica.com' → 'Bank of America'

# Account number in email
'Account: 334224' → Account ID extracted

Step 2: Contact Extraction¶

Email Signature Parsing:

Ganesh Kasthurirangan
VP/Senior Manager of Software Engineering
IP Foundational Services - Network Services
Phone: 614.209-2237
[email protected]

↓ Extracted:
- Name: Ganesh Kasthurirangan
- Title: VP/Senior Manager of Software Engineering  
- Role: decision_maker (VP detected)
- Organization: IP Foundational Services
- Phone: 614.209-2237
- Email: [email protected]

Step 3: Issue Classification¶

Keyword-Based Classification:

Context Enhancement: - No error messages → Not technical - VP escalation → High priority - Deadline mentioned → Urgent

Step 4: Urgency Assessment¶

Deadline Detection:

# Patterns
"expires Dec 31, 2025" → 2025-12-31
"by end of quarter" → Calculate Q4 end date
"ASAP" → High urgency flag

# Calculate days remaining
today = 2025-10-30
deadline = 2025-12-31
days_remaining = 62

Urgency Indicators: - "critical", "urgent", "ASAP" → High - "cannot afford", "production down" → High - "when you can", "no rush" → Low - VP/Director escalation → +1 level - Deadline < 30 days → +1 level

Step 5: Action Recommendation¶

Decision Tree:

IF issue_type == "licensing":
    → Escalate to licensing team (James McCormick)

IF issue_type == "technical" AND urgency == "high":
    → Begin troubleshooting immediately

IF issue_type == "guidance":
    → Provide documentation/consultation

IF deadline < 30 days AND business_impact == "high":
    → Add to high-priority queue

Confidence Scoring¶

Each extracted field has a confidence score:

Example:

case_number: "04293185"
confidence: HIGH (95%)
reasoning: "Explicit 'case# 04293185' in email"

customer: "JP Morgan Chase"
confidence: HIGH (92%)
reasoning: "Email domain @jpmchase.com"

issue_type: "licensing"
confidence: HIGH (89%)
reasoning: "Keywords: subscription, renewal, licenses"

contacts:
  - name: "Ganesh Kasthurirangan"
    confidence: HIGH (98%)
    reasoning: "Parsed from email signature"

Feature Status: Current vs. Roadmap¶

✅ Available Now (v2.1.0)¶

Core Intelligence: - ✅ Pattern-based email analysis - ✅ Case number extraction (95% accuracy) - ✅ Customer identification (92% accuracy) - ✅ Issue classification (89% accuracy) - ✅ Contact extraction with role detection - ✅ Urgency assessment with deadline detection - ✅ Action recommendations - ✅ Confidence scoring

Data Management: - ✅ SQLite database persistence - ✅ Case intelligence history - ✅ Feedback recording system - ✅ Database health checks

Deployment: - ✅ Container deployment (Linux) - ✅ AppImage (Linux) - ✅ DMG installer (macOS) - ✅ EXE installer (Windows) - ✅ Systemd service integration

Integrations: - ✅ JIRA API integration - ✅ Customer Portal API - ✅ rhcase bot integration - ✅ Red Hat SSO

🚧 Beta Features (v2.1.x)¶

Features available but being refined based on user feedback:

• 🚧 Team Intelligence Sharing - Share patterns across team members
• 🚧 Custom Pattern Editor - Create custom analysis rules
• 🚧 Export/Import - Backup and restore intelligence data

📋 Roadmap (Future Releases)¶

Phase 1: Advanced Pattern Learning (Q1 2026) - 📋 Machine learning from TAM feedback - 📋 Automatic pattern optimization - 📋 Team-wide accuracy tracking - 📋 A/B testing for patterns

Phase 2: Team Intelligence (Q2 2026) - 📋 Centralized pattern library - 📋 Collaborative pattern development - 📋 Team-wide statistics dashboard - 📋 Best practices sharing

Phase 3: Predictive Analysis (Q3 2026) - 📋 Issue escalation prediction - 📋 Customer health scoring - 📋 Proactive case recommendations - 📋 Risk assessment automation

Phase 4: Ansai AI Integration (Q4 2026) - 📋 Fabric pattern generation (for internal TAM docs only) - 📋 LiteLLM integration (for training materials only) - 📋 Advanced analytics (non-customer data only) - 📋 AI-assisted documentation

Accuracy Metrics¶

Current Performance¶

Accuracy by Issue Type¶

Continuous Improvement¶

graph LR
    A[Analysis] --> B[TAM Feedback]
    B --> C[Pattern Refinement]
    C --> D[Improved Accuracy]
    D --> A

Feedback System: 1. TAM reviews intelligence results 2. Marks correct/incorrect extractions 3. Patterns automatically updated 4. Next analysis more accurate

Technical Implementation¶

Database Schema¶

Case Intelligence Storage:

CREATE TABLE case_intelligence (
    id INTEGER PRIMARY KEY,
    case_number TEXT,
    customer_name TEXT,
    issue_type TEXT,
    urgency_level TEXT,
    deadline DATE,
    confidence_score REAL,
    analysis_date TIMESTAMP,
    raw_email TEXT,
    feedback_correct BOOLEAN
);

Pattern Configuration¶

Extensible Pattern System:

# config/patterns.yaml
case_patterns:
  - pattern: 'case[#\s]*(\d{8})'
    confidence: 0.95
  - pattern: 'case\s+number[:\s]*(\d{8})'
    confidence: 0.90

customer_mappings:
  '@jpmchase.com': 
    name: 'JP Morgan Chase'
    account: '334224'
    tier: 'strategic'

API Interface¶

# Intelligence Engine API
from taminator.core import IntelligenceEngine

engine = IntelligenceEngine()
result = engine.analyze_email(email_text)

print(result.case_number)        # "04293185"
print(result.confidence_score)   # 0.89
print(result.recommended_actions) # ["escalate_to_licensing"]

Best Practices for TAMs¶

1. Provide Complete Email Context¶

• Include full email thread
• Keep signatures intact
• Don't remove headers

2. Review Confidence Scores¶

• HIGH confidence → Trust it
• MEDIUM confidence → Verify
• LOW confidence → Clarify with customer

3. Give Feedback¶

• Mark correct/incorrect extractions
• Helps improve accuracy
• Benefits entire TAM team

4. Use for Initial Triage¶

• Quick case assessment
• Priority determination
• Initial routing decisions

5. Combine with Human Judgment¶

• Intelligence augments, doesn't replace
• TAM expertise is critical
• Use for speed, not blindly

Comparison: Taminator vs. Manual Analysis¶

Quality Improvement: - Consistent classification - No missed deadlines - Fewer routing errors - Better documentation

Future: Ansai Intelligence Integration¶

Taminator will integrate ansai's advanced AI capabilities while maintaining Red Hat compliance:

Internal TAM Tools (Non-Customer Data)¶

• Fabric Patterns: Generate TAM documentation
• LiteLLM: Analyze training materials
• AI Workflows: Automate internal processes

Customer Data (Offline Only)¶

• Pattern Matching: Continue current approach
• No External APIs: Maintain compliance
• Local Processing: Everything stays local

Hybrid Approach¶

Customer Email → Taminator (Offline Patterns)
Internal Notes → Ansai + Fabric (Advanced AI)
Training Docs → LiteLLM (Red Hat Models)

Next Steps¶

• 📖 Pattern Matching Details
• 📊 Accuracy Metrics
• 🎓 Learning System
• ⚙️ Customization Guide

Philosophy: Intelligence that augments TAMs, doesn't replace them.

Powered by pattern-based AI. Red Hat compliant. TAM approved.