Natural Language Processing

Learn how SmartOps understands context and intent through natural language processing, fuzzy matching, and conversational AI.

Natural Language Processing Features

SmartOps uses Amazon Bedrock’s Claude AI to understand context and intent. Examples:

Context Awareness

$$User: "Show me expensive instances" Bot: [Lists instances sorted by cost] User: "Which of those can be rightsized?" Bot: [Understands "those" refers to expensive instances]$$

Intent Recognition

$$User: "My website is down" Bot: "I'll help troubleshoot. Let me check your web servers..." [Automatically identifies web-tagged instances and checks health]$$

Fuzzy Matching

$$User: "Check the databse server" Bot: "Checking database-server-01..." [Handles typos and variations]$$

How SmartOps Understands You

1. Intent Classification

SmartOps recognizes different types of requests:

Information Requests:

  • “What instances do I have?”
  • “Show me the current status”
  • “How much am I spending?”

Action Requests:

  • “Restart the web server”
  • “Stop the test instances”
  • “Update all development servers”

Troubleshooting Requests:

  • “My application is slow”
  • “Why is the database not responding?”
  • “Help me fix this error”

2. Context Tracking

SmartOps remembers conversation context:

Example Conversation:

$$User: "List my production instances" Bot: [Shows 5 production instances] User: "Which one has the highest CPU?" Bot: "Among your production instances, web-prod-02 has the highest CPU at 78%" User: "Show me more details about that one" Bot: [Shows detailed info for web-prod-02] User: "Can you help me optimize it?" Bot: "I can help optimize web-prod-02. Let me analyze its usage patterns..."$$

3. Entity Recognition

SmartOps identifies specific entities in your requests:

Instance References:

  • Instance IDs: “i-0abc123def”
  • Instance names: “web-server-01”
  • Tags: “all production instances”
  • Roles: “database servers”, “web servers”

Time References:

  • “last week”, “yesterday”, “this month”
  • “since 9 AM”, “in the past hour”

Metric References:

  • “high CPU”, “low memory”, “disk space”
  • “expensive instances”, “underutilized servers”

Conversation Patterns

Follow-up Questions

After any response, you can ask natural follow-ups:

After Health Report:

  • “What’s causing the high CPU on server-01?”
  • “How can I fix the memory issue?”
  • “Should I be worried about that warning?”

After Cost Analysis:

  • “How much would rightsizing save me?”
  • “Which instances should I prioritize?”
  • “Can you help me implement those recommendations?”

After Instance List:

  • “Which ones need updates?”
  • “Show me only the problematic ones”
  • “What would you recommend for optimization?”

Conversational Shortcuts

SmartOps understands casual language:

Informal Requests:

  • “What’s up with my servers?” → Health status check
  • “How are things looking?” → Overall status report
  • “Any issues I should know about?” → Problem identification
  • “Can you help me save money?” → Cost optimization analysis

Implicit Context:

  • “Restart it” (after discussing a specific instance)
  • “Show me more” (after any report or list)
  • “What do you think?” (asking for AI recommendations)

Advanced Language Features

Comparative Queries

$$User: "Which instances use more CPU than database-01?" User: "Show me servers that cost more than $100/month" User: "Find instances smaller than t3.medium"$$

Temporal Queries

$$User: "What happened to web-server-01 yesterday?" User: "Show me cost trends over the last 3 months" User: "Which instances were restarted this week?"$$

Conditional Logic

$$User: "If the CPU is over 80%, show me the top processes" User: "When database connections exceed 50, check memory usage" User: "Alert me if any production instance goes down"$$

Aggregation Requests

$$User: "Total cost of all development instances" User: "Average CPU across web servers" User: "Count of instances by environment"$$

Handling Ambiguity

When your request is unclear, SmartOps will ask for clarification:

Example Ambiguous Requests:

$$User: "Restart the server" Bot: "I found 5 servers. Which one would you like me to restart? - web-server-01 (production) - web-server-02 (production) - app-server-01 (staging) - db-server-01 (production) - test-server-01 (development)" User: "Stop the expensive instances" Bot: "I found 3 instances with costs over $200/month. Would you like to: A) See details about these instances first B) Stop all 3 instances C) Stop only non-production instances"$$

Error Recovery

SmartOps handles common communication issues:

Typos and Misspellings

$$User: "chck the helth of web-srvr-01" Bot: "Checking the health of web-server-01..." [Automatically corrects and understands intent]$$

Incomplete Requests

$$User: "show instances" Bot: "Here are your EC2 instances. Would you like me to: - Show more details about any specific instance - Filter by environment or status - Include cost or performance information"$$

Mixed Languages (Limited)

$$User: "montrer les instances" Bot: "I understand you want to see instances. Here are your EC2 instances..." [Basic French recognition, responds in English]$$

Learning from Context

SmartOps learns patterns within a conversation:

Preference Learning

$$User: "Show instances without IDs" Bot: [Remembers preference for this conversation] User: "Now show the database servers" Bot: [Lists database servers without IDs, as previously requested]$$

Domain Knowledge

SmartOps understands infrastructure terminology:

Technical Terms:

  • “Load balancer”, “auto-scaling group”, “availability zone”
  • “RDS”, “EBS”, “VPC”, “security group”
  • “Production”, “staging”, “development” environments

Common Patterns:

  • “Scale up/down”, “failover”, “backup”
  • “Peak hours”, “maintenance window”, “scheduled downtime”
  • “Performance bottleneck”, “capacity planning”

Best Practices for Natural Communication

1. Be Conversational

  • Use natural language like you’re talking to a colleague
  • Don’t worry about exact command syntax
  • Feel free to ask follow-up questions

2. Provide Context

  • Mention environment: “production servers”, “dev instances”
  • Include timeframes: “since yesterday”, “this week”
  • Specify scope: “all web servers”, “just the database”

3. Ask for Clarification

  • If responses aren’t what you expected, ask for clarification
  • Request more details: “Can you explain that further?”
  • Ask for alternatives: “What other options do I have?”

4. Build on Previous Responses

  • Use “that instance”, “those servers”, “the expensive ones”
  • Ask follow-ups: “What would you recommend?”, “Should I be concerned?”

Integration with Commands

Natural language seamlessly integrates with command execution:

$$User: "The production web servers seem slow, can you help?" Bot: "I'll investigate your production web servers. Let me check their current status... Found 3 production web servers: - web-prod-01: CPU 45%, Memory 78% - web-prod-02: CPU 89%, Memory 92% ⚠️ - web-prod-03: CPU 52%, Memory 65% web-prod-02 appears to have high resource usage. Would you like me to: A) Show detailed process information B) Restart the web services C) Investigate recent logs D) Recommend instance scaling"$$

Next Steps

Need Help?