This guide provides detailed instructions for creating and configuring AI personas in Psyche. Learn how to craft effective personalities, customize voices, tune behaviors, and optimize your virtual assistants for specific use cases.
Table of Contents#
- Understanding Personas
- Planning Your Persona
- Identity Configuration
- Personality Engineering
- Expertise & Knowledge
- Voice Customization
- Avatar Configuration
- Behavior Tuning
- Memory & Context
- Testing & Iteration
- Use Case Examples
- Best Practices
- Troubleshooting
Understanding Personas#
A persona is the complete configuration of an AI virtual assistant, defining:
- Who they are (identity)
- How they communicate (personality)
- What they know (expertise)
- How they sound (voice)
- How they look (avatar)
- What they remember (memory)
Persona Components#
┌─────────────────────────────────────────────────────────────┐
│ PERSONA │
├──────────────┬──────────────┬──────────────┬───────────────┤
│ Identity │ Personality │ Expertise │ Voice │
│ ──────────── │ ──────────── │ ──────────── │ ──────────── │
│ • Name │ • OCEAN │ • Domains │ • Provider │
│ • Title │ • Style │ • Skills │ • Voice ID │
│ • Company │ • Tone │ • Levels │ • Speed │
│ • Bio │ • Rules │ • Knowledge │ • Pitch │
├──────────────┼──────────────┴──────────────┴───────────────┤
│ Avatar │ Memory & Context │
│ ──────────── │ ─────────────────────────────────────────── │
│ • Style │ • Facts • Opinions • Conversation │
│ • Quality │ • History • Preferences • Session State │
└──────────────┴─────────────────────────────────────────────┘
Why Persona Design Matters#
Well-designed personas:
- Create consistent, predictable interactions
- Build trust with users through authentic behavior
- Improve task completion rates
- Reduce user confusion and frustration
- Enhance brand representation
Planning Your Persona#
Before creating a persona, answer these key questions:
1. Define the Purpose#
| Question | Why It Matters |
|---|---|
| What tasks will this persona handle? | Determines required expertise |
| Who will interact with it? | Influences communication style |
| What outcomes should it drive? | Shapes behavior and responses |
| What brand does it represent? | Guides personality and appearance |
2. Identify Target Users#
| User Type | Recommended Approach |
|---|---|
| Enterprise B2B | Professional, formal, domain expert |
| Consumer B2C | Friendly, accessible, helpful |
| Technical users | Precise, detailed, technical vocabulary |
| General public | Simple language, patient, educational |
| Internal teams | Efficient, direct, context-aware |
3. Map to Business Goals#
| Goal | Persona Characteristics |
|---|---|
| Customer support | Empathetic, patient, solution-focused |
| Sales | Persuasive, knowledgeable, engaging |
| Training | Clear, structured, encouraging |
| Brand representation | Consistent, memorable, on-brand |
| Technical support | Precise, thorough, methodical |
Planning Template#
Persona Name: _______________________
Primary Role: _______________________
Target Audience: ____________________
Key Goals:
1. _______________________________
2. _______________________________
3. _______________________________
Success Metrics:
- ________________________________
- ________________________________
Brand Alignment: ____________________
Identity Configuration#
Identity forms the foundation of your persona.
Required Fields#
Name#
Guidelines:
- Use realistic first and last names
- Match cultural expectations of your audience
- Consider pronunciation in voice synthesis
- Avoid trademarked names
Examples: | Use Case | Good Names | Avoid | | ------------------ | ----------------------------- | ------------------ | | Professional | Alex Chen, Sarah Williams | Bot, Helper | | Friendly | Jamie, Taylor, Morgan | Assistant123 | | Technical | Dr. Marcus Webb, Prof. Singh | TechBot | | Creative | Luna, Phoenix, River | Generic names |
Job Title#
Guidelines:
- Reflect actual expertise level
- Use industry-standard titles
- Consider seniority implications
Examples:
Customer Success Manager
Senior Sales Consultant
Technical Support Engineer
Product Specialist
Training Coordinator
Bio/Description#
Guidelines:
- Keep to 2-3 sentences
- Include relevant background
- Establish credibility
- Make it memorable
Template:
[Name] is a [title] with [X years] of experience in [domain].
[He/She/They] specialize[s] in [specific areas] and [has/have]
helped [achievement/client type]. [Personal touch or philosophy].
Example:
Alex is a Senior Customer Success Manager with 8 years of experience
in enterprise software. She specializes in helping teams optimize their
workflows and has guided over 200 companies through successful
implementations. Alex believes that great support is about understanding
the person, not just the problem.
Optional Fields#
Company/Organization#
Link the persona to a specific organization:
Company: TechCorp Solutions
Division: Customer Success Team
Location: San Francisco, CA
Profile Photo#
Upload requirements:
- Format: JPEG or PNG
- Size: 400x400 pixels minimum
- Style: Professional headshot or avatar illustration
- Tip: Ensure it matches the avatar if using video
Personality Engineering#
Personality determines how your persona communicates.
The OCEAN Model (Big Five)#
Psyche uses the scientifically-validated Big Five personality model:
Openness (0.0 - 1.0)#
| Score | Characteristics |
|---|---|
| Low | Conventional, practical, prefers routine |
| Mid | Balanced curiosity, adaptable |
| High | Creative, curious, enjoys novel ideas |
When to use high openness:
- Creative industries
- Innovation-focused roles
- Brainstorming sessions
When to use low openness:
- Compliance-focused roles
- Standardized processes
- Risk-averse contexts
Conscientiousness (0.0 - 1.0)#
| Score | Characteristics |
|---|---|
| Low | Flexible, spontaneous, adaptable |
| Mid | Balanced organization, reasonable attention |
| High | Organized, thorough, detail-oriented |
When to use high conscientiousness:
- Technical support
- Legal/compliance
- Project management
- Financial services
When to use low conscientiousness:
- Creative brainstorming
- Casual customer chat
- Entertainment
Extraversion (0.0 - 1.0)#
| Score | Characteristics |
|---|---|
| Low | Reserved, thoughtful, calm |
| Mid | Situationally social, balanced energy |
| High | Outgoing, energetic, enthusiastic |
When to use high extraversion:
- Sales
- Event hosting
- Team motivation
- Customer engagement
When to use low extraversion:
- Technical documentation
- Research assistance
- Counseling/support
Agreeableness (0.0 - 1.0)#
| Score | Characteristics |
|---|---|
| Low | Direct, challenging, competitive |
| Mid | Balanced cooperation and assertion |
| High | Cooperative, warm, trusting |
When to use high agreeableness:
- Customer support
- Healthcare
- Education
- Conflict resolution
When to use low agreeableness:
- Debate training
- Competitive sales
- Critical feedback
Neuroticism (0.0 - 1.0)#
| Score | Characteristics |
|---|---|
| Low | Emotionally stable, calm under pressure |
| Mid | Normal emotional range |
| High | Sensitive, empathetic to stress |
Recommendation: Keep neuroticism low (0.1-0.3) for most use cases. High neuroticism can make responses seem anxious or uncertain.
Communication Style#
Set the overall communication approach:
| Style | Description | Best For |
|---|---|---|
| Formal | Professional, structured, reserved | Legal, finance |
| Professional | Business-appropriate, clear | B2B, enterprise |
| Casual | Relaxed, conversational | B2C, retail |
| Friendly | Warm, approachable, personable | Support, education |
| Technical | Precise, jargon-appropriate | IT, engineering |
Conversational Tone#
Define the emotional quality of interactions:
| Tone | Characteristics | Use Cases |
|---|---|---|
| Warm | Caring, supportive, encouraging | Support, coaching |
| Neutral | Balanced, informational, objective | Documentation |
| Enthusiastic | Energetic, positive, motivating | Sales, events |
| Calm | Relaxed, measured, reassuring | Healthcare, crisis |
| Assertive | Confident, direct, action-oriented | Training, leadership |
Custom Instructions#
Add specific behavioral rules:
# Communication Rules
- Always greet users by name when available
- Ask clarifying questions before providing solutions
- Use analogies to explain complex concepts
- Avoid technical jargon unless the user demonstrates expertise
- End conversations with a satisfaction check
# Boundaries
- Never discuss competitor products negatively
- Redirect political discussions politely
- Escalate to human support after 3 failed resolution attempts
- Do not make promises about features or timelines
# Formatting
- Use bullet points for lists
- Keep responses under 150 words unless detail is requested
- Include relevant links to documentation
- Use code blocks for technical examples
Personality Presets#
Quick-start with predefined personality configurations:
Warm Professional#
openness: 0.6
conscientiousness: 0.8
extraversion: 0.6
agreeableness: 0.8
neuroticism: 0.2
style: Professional
tone: Warm
Best for: Customer success, account management, onboarding
Analytical Expert#
openness: 0.8
conscientiousness: 0.9
extraversion: 0.4
agreeableness: 0.6
neuroticism: 0.3
style: Technical
tone: Neutral
Best for: Technical support, data analysis, research assistance
Friendly Guide#
openness: 0.7
conscientiousness: 0.6
extraversion: 0.8
agreeableness: 0.9
neuroticism: 0.2
style: Casual
tone: Enthusiastic
Best for: Retail, hospitality, entertainment, general consumer
Calm Advisor#
openness: 0.5
conscientiousness: 0.7
extraversion: 0.4
agreeableness: 0.8
neuroticism: 0.1
style: Professional
tone: Calm
Best for: Healthcare, financial planning, crisis support
Expertise & Knowledge#
Define what your persona knows and how deeply.
Expertise Domains#
Add multiple domains with proficiency levels:
| Level | Description |
|---|---|
| Beginner | Basic understanding, can answer simple questions |
| Intermediate | Working knowledge, handles common scenarios |
| Advanced | Deep expertise, handles complex cases |
| Expert | Authority-level, handles edge cases |
Domain Configuration#
expertise:
- domain: 'Customer Support'
level: Expert
topics:
- Ticket handling
- Escalation procedures
- Customer satisfaction
- domain: 'Product Knowledge'
level: Expert
topics:
- Feature documentation
- Use cases
- Troubleshooting
- domain: 'Technical Integration'
level: Advanced
topics:
- API usage
- SDK implementation
- Webhooks
Linking Knowledge Base#
Connect personas to relevant knowledge collections:
knowledge_base:
collections:
- id: 'kb_product_docs'
priority: high
- id: 'kb_faq'
priority: high
- id: 'kb_technical_guides'
priority: medium
search_config:
top_k: 5
min_score: 0.7
Expertise Boundaries#
Define what the persona should NOT attempt:
boundaries:
redirect_topics:
- "Legal advice" → "Please consult our legal team"
- "Medical diagnosis" → "Please consult a healthcare provider"
- "Financial advice" → "Please consult a financial advisor"
escalation_triggers:
- Repeated failed attempts (3+)
- User frustration detected
- Out-of-scope requests
- Sensitive personal data
Voice Customization#
Configure text-to-speech for natural spoken interactions.
Voice Provider Selection#
| Provider | Strengths | Best For |
|---|---|---|
| ElevenLabs | Most natural, emotion range | Customer-facing |
| Google TTS | Wide language support | Multilingual |
| Amazon Polly | Cost-effective, reliable | High volume |
| Azure | Enterprise integration | Microsoft environments |
Voice Selection#
Choosing a Voice#
Consider these factors:
| Factor | Guidelines |
|---|---|
| Gender | Match persona identity or stay neutral |
| Age | Match perceived persona age |
| Accent | Match target audience expectations |
| Energy | Match persona personality (calm vs energetic) |
| Clarity | Prioritize for technical content |
Popular ElevenLabs Voices#
| Voice | Characteristics | Good For |
|---|---|---|
| Rachel | Professional female, clear | Business, support |
| Josh | Friendly male, warm | Sales, onboarding |
| Emily | Energetic female, engaging | Marketing, events |
| Sam | Calm male, reassuring | Healthcare, finance |
| Bella | Youthful female, conversational | Consumer, retail |
| Adam | Authoritative male, confident | Training, leadership |
Voice Parameters#
Speed (0.5x - 2.0x)#
| Speed | Use Case |
|---|---|
| 0.7x | Complex explanations, accessibility |
| 0.85x | Technical content, older audiences |
| 1.0x | Standard conversation (default) |
| 1.15x | Casual chat, younger audiences |
| 1.3x | Quick updates, familiar users |
Pitch (-50% to +50%)#
| Adjustment | Effect |
|---|---|
| -20% | Deeper, more authoritative |
| -10% | Slightly deeper, calm |
| 0% | Natural voice (default) |
| +10% | Slightly higher, friendly |
| +20% | Higher, energetic |
Voice Stability & Clarity#
For ElevenLabs:
voice_settings:
stability: 0.75 # 0=variable, 1=consistent
similarity_boost: 0.75 # 0=diverse, 1=close to original
style: 0.0 # 0=neutral, 1=expressive
use_speaker_boost: true
Recommendations:
| Context | Stability | Similarity |
|---|---|---|
| Professional | 0.8 | 0.8 |
| Conversational | 0.6 | 0.7 |
| Emotional support | 0.5 | 0.6 |
| Technical reading | 0.9 | 0.9 |
Voice Cloning#
Create a custom voice from audio samples:
Requirements#
- 1-3 minutes of clear speech
- Single speaker only
- Minimal background noise
- Consistent audio quality
- Natural speaking pace
Process#
- Record or collect audio samples
- Upload to voice provider
- Wait for processing (typically 5-30 minutes)
- Preview and adjust settings
- Test in context before deployment
Avatar Configuration#
Configure visual representation for video sessions.
Avatar Styles#
| Style | Description | Use Case |
|---|---|---|
| Photorealistic | Lifelike human appearance | Customer-facing |
| Animated | Stylized cartoon character | Casual, internal |
| Stylized | Artistic interpretation | Brand-specific |
| None | Audio only, no video | Voice-only sessions |
Quality Settings#
| Quality | Resolution | FPS | Latency | GPU |
|---|---|---|---|---|
| 480p | 854x480 | 24 | <20ms | Minimal |
| 720p | 1280x720 | 30 | <25ms | Standard |
| 1080p | 1920x1080 | 30 | <30ms | High |
| 4K | 3840x2160 | 30 | <40ms | Very High |
Recommendation: Use 720p for most deployments; 1080p for executive/demo scenarios.
Training Custom Avatars#
Video Recording Guidelines#
| Requirement | Specification |
|---|---|
| Duration | 30-60 seconds |
| Resolution | 1080p minimum (4K preferred) |
| Frame rate | 30 FPS minimum |
| Lighting | Even, soft, diffused (no harsh shadows) |
| Background | Solid color, preferably green/gray |
| Framing | Head and shoulders, centered |
| Movement | Slow head rotations covering all angles |
| Expression | Neutral, smile, talking (variety) |
Recording Checklist#
□ Clean camera lens
□ Stable camera/tripod
□ Even lighting from multiple angles
□ Solid background (no patterns)
□ Subject centered in frame
□ Head and shoulders visible
□ Minimal makeup/accessories
□ Remove glasses if possible
□ Wear solid, non-reflective clothing
□ Record in quiet environment
Training Process#
- Upload video to Psyche
- System extracts face geometry
- FLAME model fit (parametric face model)
- 3D Gaussian Splatting training (~30,000 iterations)
- Expression blendshape learning
- Quality validation
- Avatar ready for deployment
Typical Training Time: 1-4 hours depending on video quality
Avatar Expressions#
The avatar automatically animates based on:
- Speech content and emotion
- FLAME expression parameters (50 blend shapes)
- Lip sync from TTS output
Expression Mapping#
| Emotion | Automatic Trigger |
|---|---|
| Neutral | Default state |
| Happy | Positive language, greetings |
| Concerned | Problem acknowledgment, empathy |
| Thoughtful | Questions, complex explanations |
| Surprise | Unexpected input, exclamations |
Behavior Tuning#
Fine-tune how your persona responds in different situations.
Response Length#
response_behavior:
default_length: 'medium' # short, medium, long
length_rules:
- context: 'quick_answer'
length: 'short'
max_words: 50
- context: 'explanation'
length: 'medium'
max_words: 150
- context: 'detailed_guide'
length: 'long'
max_words: 500
Response Formatting#
formatting:
use_markdown: true
bullet_points: true
numbered_lists: true
code_blocks: true
max_paragraph_length: 3 # sentences
preferences:
- Prefer bullet points for lists of 3+
- Use headers for long responses
- Include code examples for technical topics
- Add relevant links when available
Conversation Flow#
Turn-Taking#
conversation:
wait_for_completion: true # Wait for user to finish speaking
acknowledgment_phrases:
- 'I understand'
- 'Got it'
- 'Let me help with that'
clarification_behavior:
threshold: 0.7 # Confidence below this triggers clarification
phrases:
- 'Just to make sure I understand...'
- 'Could you clarify...'
- 'Did you mean...'
Topic Transitions#
transitions:
smooth_topic_change: true
phrases:
new_topic: 'Moving on to...'
return_topic: 'Going back to what you mentioned...'
closing: 'Is there anything else I can help with?'
Error Handling#
error_handling:
unknown_query:
response: "I'm not sure about that. Let me find someone who can help."
action: escalate
repeated_failure:
threshold: 3
response:
"I apologize that I haven't been able to help. Let me connect you with a
specialist."
action: escalate
out_of_scope:
response:
"That's outside my area of expertise. I can connect you with [department]."
action: redirect
Proactive Behaviors#
proactive:
suggestions:
enabled: true
trigger: 'after_resolution'
examples:
- 'Would you like me to send you our guide on this topic?'
- 'I can also help you with [related task]. Would that be useful?'
follow_ups:
enabled: true
timing: 'before_close'
examples:
- 'Did that solve your issue?'
- "Is there anything else you'd like to know?"
Memory & Context#
Configure what your persona remembers across interactions.
Fact Memory#
Store persistent information about the persona:
facts:
- key: 'company_name'
value: 'TechCorp Solutions'
category: 'organization'
- key: 'support_hours'
value: 'Monday-Friday, 9am-6pm EST'
category: 'policy'
- key: 'refund_policy'
value: '30-day money-back guarantee'
category: 'policy'
- key: 'current_promotion'
value: '20% off annual plans'
category: 'sales'
Opinion Memory#
Define the persona's stance on topics:
opinions:
- topic: 'remote_work'
stance:
'I believe remote work can be highly effective with the right tools and
communication.'
strength: 0.8 # How strongly held
- topic: 'our_product'
stance:
'Our solution is designed to make teams more productive while being easy
to use.'
strength: 0.9
- topic: 'ai_ethics'
stance: 'AI should augment human capabilities, not replace human judgment.'
strength: 0.7
Session Context#
Information that persists within a session:
session_context:
remember:
- user_name
- stated_preferences
- current_issue
- previous_solutions_tried
- sentiment_trajectory
forget_after_session:
- personal_details
- sensitive_information
Cross-Session Memory#
For returning users (requires user identification):
user_memory:
enabled: true
retention_days: 90
remember:
- interaction_history
- stated_preferences
- unresolved_issues
- product_usage
privacy:
anonymize_after_days: 365
user_can_delete: true
Testing & Iteration#
Validate and improve your persona before deployment.
Pre-Launch Testing#
Conversation Testing#
Test common scenarios:
□ Greeting and introduction
□ Simple FAQ responses
□ Complex multi-turn conversations
□ Edge cases and unusual requests
□ Error handling
□ Escalation triggers
□ Session ending
Personality Verification#
| Test | Method |
|---|---|
| Consistency | Ask same question multiple ways |
| Tone adherence | Check emotional situations |
| Knowledge boundaries | Ask out-of-scope questions |
| Brand alignment | Verify language matches guidelines |
Voice & Avatar Quality#
□ Voice naturalness in long responses
□ Pronunciation of domain terms
□ Lip sync accuracy
□ Expression appropriateness
□ Latency under load
Metrics to Track#
| Metric | Target | Description |
|---|---|---|
| Task completion rate | > 80% | Issues resolved without escalation |
| Satisfaction score | > 4.0/5 | Post-interaction rating |
| Avg response time | < 500ms | Time to first response |
| Conversation length | Optimal | Neither too short nor too long |
| Escalation rate | < 20% | Conversations requiring human help |
Iteration Process#
-
Collect Feedback
- Review conversation transcripts
- Analyze satisfaction scores
- Identify escalation patterns
-
Identify Issues
- Knowledge gaps
- Personality mismatches
- Voice/avatar problems
- Behavior inconsistencies
-
Make Adjustments
- Update system prompt
- Adjust personality parameters
- Add knowledge base content
- Tune voice settings
-
Validate Changes
- A/B test with subset of users
- Compare metrics before/after
- Get qualitative feedback
-
Deploy Updates
- Roll out changes gradually
- Monitor for regressions
- Document changes
Use Case Examples#
Customer Support Agent#
name: 'Sarah Chen'
title: 'Customer Success Manager'
company: 'TechCorp'
personality:
openness: 0.6
conscientiousness: 0.8
extraversion: 0.6
agreeableness: 0.9
neuroticism: 0.2
style: Professional
tone: Warm
expertise:
- domain: 'Product Support'
level: Expert
- domain: 'Account Management'
level: Expert
- domain: 'Technical Troubleshooting'
level: Advanced
custom_instructions: |
- Always acknowledge the customer's issue before providing solutions
- Ask clarifying questions to understand the full context
- Provide step-by-step guidance when explaining processes
- Offer to follow up after complex issues are resolved
- Escalate to human support if unable to resolve after 3 attempts
voice:
provider: 'elevenlabs'
voice_id: 'rachel'
speed: 1.0
pitch: 0
Sales Development Representative#
name: 'Marcus Wright'
title: 'Senior Sales Consultant'
company: 'TechCorp'
personality:
openness: 0.7
conscientiousness: 0.7
extraversion: 0.8
agreeableness: 0.7
neuroticism: 0.2
style: Professional
tone: Enthusiastic
expertise:
- domain: 'Sales'
level: Expert
- domain: 'Product Demos'
level: Expert
- domain: 'Pricing & Packages'
level: Advanced
- domain: 'Competitive Analysis'
level: Intermediate
custom_instructions: |
- Focus on understanding the prospect's business challenges first
- Highlight relevant features that address their specific needs
- Use success stories and case studies when appropriate
- Guide towards scheduling a demo or meeting with account executive
- Never pressure or use high-pressure tactics
- Be transparent about pricing when asked
voice:
provider: 'elevenlabs'
voice_id: 'josh'
speed: 1.05
pitch: 0
Technical Support Engineer#
name: 'Dr. Aisha Patel'
title: 'Technical Support Engineer'
company: 'TechCorp'
personality:
openness: 0.8
conscientiousness: 0.9
extraversion: 0.4
agreeableness: 0.6
neuroticism: 0.2
style: Technical
tone: Neutral
expertise:
- domain: 'API Integration'
level: Expert
- domain: 'SDK Development'
level: Expert
- domain: 'Infrastructure'
level: Advanced
- domain: 'Security'
level: Advanced
custom_instructions: |
- Assume technical competence unless user indicates otherwise
- Provide code examples when relevant
- Include links to documentation
- Explain the "why" behind recommendations
- Ask for error logs and configuration details
- Be precise with terminology
voice:
provider: 'elevenlabs'
voice_id: 'sam'
speed: 0.95
pitch: -5
Training Instructor#
name: 'Jamie Taylor'
title: 'Product Training Specialist'
company: 'TechCorp'
personality:
openness: 0.7
conscientiousness: 0.7
extraversion: 0.7
agreeableness: 0.8
neuroticism: 0.2
style: Friendly
tone: Enthusiastic
expertise:
- domain: 'Product Training'
level: Expert
- domain: 'Onboarding'
level: Expert
- domain: 'Best Practices'
level: Expert
custom_instructions: |
- Start with learning objectives for each session
- Break complex topics into digestible steps
- Use analogies and real-world examples
- Check understanding frequently with questions
- Celebrate progress and achievements
- Provide practice exercises when appropriate
- Summarize key points at the end
voice:
provider: 'elevenlabs'
voice_id: 'emily'
speed: 1.0
pitch: +5
Best Practices#
Do's#
-
Start Simple
- Begin with a focused use case
- Add complexity gradually
- Iterate based on real usage
-
Be Authentic
- Create believable backstories
- Maintain consistent personality
- Avoid uncanny behavior
-
Set Clear Boundaries
- Define what the persona should NOT do
- Establish escalation paths
- Be transparent about AI nature
-
Test Thoroughly
- Test edge cases
- Verify voice quality
- Check avatar expressions
- Validate knowledge accuracy
-
Iterate Continuously
- Monitor conversation quality
- Update knowledge regularly
- Refine based on feedback
Don'ts#
-
Don't Over-Promise
- Avoid claims of human-like intelligence
- Don't guarantee specific outcomes
- Be honest about limitations
-
Don't Ignore Context
- Consider cultural differences
- Account for accessibility needs
- Adapt to user preferences
-
Don't Skip Testing
- Always test before deployment
- Include edge cases
- Verify quality at scale
-
Don't Set and Forget
- Personas need ongoing maintenance
- Knowledge becomes outdated
- User needs evolve
-
Don't Overcomplicate
- Simple personas often work best
- Avoid conflicting personality traits
- Focus on core use cases
Troubleshooting#
Personality Issues#
Persona feels inconsistent#
Symptoms: Responses vary widely in tone or style
Solutions:
- Reduce openness to add consistency
- Add more specific custom instructions
- Clarify conflicting personality traits
- Review and update system prompt
Persona is too formal/informal#
Symptoms: Communication style doesn't match expectations
Solutions:
- Adjust formality setting in communication style
- Modify extraversion (higher = more casual)
- Update custom instructions with examples
- Add sample phrases to guide style
Voice Issues#
Voice sounds robotic#
Solutions:
- Reduce stability setting (allow more variation)
- Try different voice model
- Adjust speed to more natural pace
- Enable speaker boost
Pronunciation errors#
Solutions:
- Add pronunciation hints in text
- Use phonetic spelling for unusual words
- Try different voice provider
- Add domain terms to pronunciation dictionary
Avatar Issues#
Avatar looks unnatural#
Solutions:
- Re-record training video with better lighting
- Ensure variety of expressions in training data
- Check video resolution and quality
- Try different avatar style
Expressions don't match speech#
Solutions:
- Verify emotion detection is working
- Adjust expression sensitivity settings
- Check for audio/video sync issues
- Re-train avatar with more expression variety
Knowledge Issues#
Persona gives wrong information#
Solutions:
- Update knowledge base with correct information
- Increase knowledge base priority
- Add specific facts to persona memory
- Review and correct system prompt
Persona can't answer common questions#
Solutions:
- Add relevant documents to knowledge base
- Increase expertise level for related domain
- Add FAQ content explicitly
- Review conversation logs for gaps
Appendix#
Personality Trait Reference#
| Trait | 0.0-0.3 (Low) | 0.4-0.6 (Mid) | 0.7-1.0 (High) |
|---|---|---|---|
| Openness | Practical, routine | Balanced | Creative, curious |
| Conscientiousness | Flexible, casual | Organized | Meticulous, precise |
| Extraversion | Reserved, calm | Balanced | Outgoing, energetic |
| Agreeableness | Direct, assertive | Cooperative | Warm, trusting |
| Neuroticism | Stable, calm | Normal emotional | Sensitive, anxious |
Voice Provider Comparison#
| Feature | ElevenLabs | Amazon | Azure | |
|---|---|---|---|---|
| Naturalness | Excellent | Good | Good | Good |
| Languages | 29 | 40+ | 30+ | 75+ |
| Custom voices | Yes | No | No | Yes |
| Emotion control | Yes | Limited | Limited | Limited |
| Real-time | Yes | Yes | Yes | Yes |
| Cost | $$$$ | $$ | $$ | $$ |
Common Mistakes#
| Mistake | Fix |
|---|---|
| Conflicting personality traits | Align OCEAN values with intended use |
| Too many custom instructions | Focus on key behaviors only |
| Incomplete knowledge base | Add comprehensive domain content |
| Ignoring voice quality | Test audio before deployment |
| No escalation path | Define clear handoff procedures |
| Static persona | Schedule regular reviews and updates |