Prompt Engineering Guide

Workflows

POCR AI™ offers four distinct workflow patterns, each requiring a different approach to prompt engineering:

• Sequential Refinement: A series of agents that progressively refine the output
• Divide & Conquer: Multiple agents work on separate parts of a complex task in parallel
• Creator-Critic: A creator agent produces content that critic agents evaluate and improve
• Majority Voting: Multiple agents vote on a binary issue, with supporting opinion

Understanding which workflow pattern best suits your task is the first step in effective prompt engineering.

Personas

Each agent in the workflow can be enhanced with a specialized persona that influences how it processes your prompts:

• CEO: Strategic, big-picture thinking with business orientation
• CTO: Technical strategy and implementation feasibility
• Technical Architect: Detailed technical design and specifications
• Data Analyst: Quantitative analysis and data-driven insights
• Designer: User experience and aesthetic considerations
• Sales: Commercial opportunities and market positioning
• Expert: Domain expertise and practical application knowledge
• User: Foundational understanding and simplification of complex concepts

Different personas respond best to different prompting approaches.

Agents

POCR AI™ supports connections to various AI models:

• OpenAI (GPT models)
• Anthropic (Claude models)
• Google (Gemini models)
• Meta (Llama models)
• DeepSeek
• Falcon
• Qwen
• Mistral

Each model has unique strengths and limitations that influence prompt engineering strategies.

Core Principles

Effective prompts in POCR AI™ follow these core principles:

1. Clarity: Be explicit about what you want
2. Context: Provide relevant background information
3. Structure: Organize information in a logical sequence
4. Specificity: Include details about format, length, tone, and style
5. Direction: Guide the thinking process with clear instructions

These principles apply regardless of which workflow pattern, personas, or models you select.

Prompt Components

A well-crafted prompt for POCR AI™ typically includes:

1. Task Description: What you want the system to accomplish
2. Input Data: Information for the system to process
3. Format Instructions: How you want the output structured
4. Evaluation Criteria: What quality standards to apply
5. Constraints: Any limitations or boundaries to respect
6. Examples: Illustrations of desired outputs (when applicable)

Not all components are necessary for every prompt, but considering each helps ensure completeness.

Context Control

Managing context is critical in multi-agent workflows:

• Explicit Context: Directly stated in your prompt
• Implicit Context: Inferred from your phrasing and previous interactions
• Persistent Context: Maintained across multiple turns in a workflow
• Context Refreshing: Reestablishing important context that might be lost during agent handoffs

For complex workflows, explicitly managing context between agents improves output coherence.

Sequential Refinement

In Sequential Refinement workflows, each agent builds upon the work of previous agents. Effective prompting requires:

• Stage-Appropriate Instructions: Tailor instructions to each stage of the process
• Progressive Detailing: Start broad and add specificity in later stages
• Context Preservation: Ensure important context isn't lost between agents
• Stage-Specific Evaluation Criteria: Define success differently at each stage

Task: Create a comprehensive marketing plan for our new product
Workflow: Sequential Refinement

Stage 1 (CEO Persona):
- Develop high-level strategic framework
- Identify target markets and positioning
- Define success metrics

Stage 2 (Sales Persona):
- Build on the strategic framework
- Develop specific marketing channels and approaches
- Create timeline and resource allocation

Stage 3 (Designer Persona):
- Refine the marketing plan with creative elements
- Enhance engagement strategies
- Add visual styling guidelines

Output Format: Comprehensive marketing plan document with executive summary, strategy section, tactical details, and implementation guidelines

Divide & Conquer

Divide & Conquer workflows distribute complex tasks across multiple agents. Effective prompting requires:

• Clear Task Decomposition: Explicitly divide the work into coherent subtasks
• Independent Subtask Definition: Ensure subtasks can be processed in parallel
• Boundary Specifications: Define how subtasks interface with each other
• Integration Guidelines: Provide instructions for recombining subtask outputs

Task: Analyze a quarterly financial report
Workflow: Divide & Conquer

Subtask 1 (Data Analyst Persona):
- Analyze key performance metrics and trends
- Identify statistical anomalies
- Create data visualization recommendations

Subtask 2 (Sales Persona):
- Evaluate market positioning
- Assess competitive landscape
- Identify revenue growth opportunities

Subtask 3 (CEO Persona):
- Analyze strategic implications
- Identify risks and opportunities
- Recommend high-level actions

Integration Instructions:
- Combine insights into a coherent analysis
- Resolve any conflicting recommendations
- Prioritize action items by impact and feasibility

Output Format: Comprehensive financial analysis report with data insights, market analysis, and strategic recommendations sections

Creator-Critic

Creator-Critic workflows involve initial content creation followed by specialized evaluation. Effective prompting requires:

• Creation Parameters: Clear guidelines for the creator agent
• Evaluation Criteria: Specific aspects for critics to assess
• Constructive Guidance: Instructions for actionable feedback
• Revision Instructions: Guidelines for incorporating feedback

Task: Develop a product feature announcement blog post
Workflow: Creator-Critic

Creator Instructions (Technical Architect Persona):
- Create a 500-word blog post announcing our new API integration feature
- Include technical capabilities, implementation details, and use cases
- Balance technical accuracy with readability

Critic 1 Instructions (CTO Persona):
- Evaluate technical accuracy and completeness
- Assess whether implementation details are correctly represented
- Identify any missing technical considerations

Critic 2 Instructions (Expert Persona):
- Evaluate clarity and understandability
- Assess whether use cases are relevant and compelling
- Identify potential questions or concerns users might have

Revision Guidelines:
- Maintain the original structure while addressing all critique points
- Prioritize technical accuracy while improving accessibility
- Add clarifying examples where appropriate

Output Format: Revised blog post with improvements based on both technical and user-centered feedback

Majority Voting

Majority Voting workflows leverage multiple agents to reach consensus. Effective prompting requires:

• Clear Decision Parameters: What question needs to be answered
• Evaluation Framework: Standards for assessing options
• Decision Options: Possible answers or approaches
• Reasoning Requirements: Instructions for justifying positions

Task: Determine the best technology framework for our new mobile application
Workflow: Majority Voting

Decision Question: Which framework should we use for our new e-commerce mobile application?

Options to Evaluate:
1. React Native
2. Flutter
3. Native iOS/Android development
4. Progressive Web App

Evaluation Criteria:
- Development efficiency
- Performance characteristics
- Long-term maintainability
- Ecosystem and community support
- Alignment with team skills

Agent Instructions (for CTO, Technical Architect, and Smart User Personas):
- Independently evaluate all options against the criteria
- Provide detailed reasoning for your recommendation
- Assign a 1-10 rating for each option with justification
- Make a final recommendation with summary rationale

Output Format: Summary of each agent's evaluation, consensus recommendation (if any), and detailed reasoning highlighting areas of agreement and disagreement

CEO

The CEO persona focuses on strategic thinking and business impact. Optimize your prompts by:

• Emphasizing Big-Picture Context: Provide market, industry, and competitive information
• Framing in Business Terms: Connect technical details to business outcomes
• Requesting Strategic Analysis: Ask for high-level insights and implications
• Including Stakeholder Considerations: Mention customers, investors, or employees

As a CEO, analyze the strategic implications of this market data. Consider:
- How does this impact our competitive position?
- What long-term opportunities or threats emerge?
- What are the key strategic decisions we should consider?
- How might this affect our relationships with key stakeholders?

Format your response with an executive summary, key strategic insights, recommended actions, and long-term considerations.

CTO

The CTO persona specializes in technical strategy and implementation feasibility. Optimize your prompts by:

• Providing Technical Context: Include system architecture and technology stack details
• Specifying Technical Constraints: Mention limitations like scalability needs or performance requirements
• Requesting Implementation Considerations: Ask for high-level technical approaches
• Including Risk Assessment: Ask for potential technical challenges and mitigations

As a CTO, evaluate the feasibility of implementing this feature into our current architecture. Our system uses a microservices approach with Python backend services and React frontend.

Consider:
- Technical integration approach
- Scalability implications
- Implementation risks and mitigations
- Resource requirements and timeline considerations
- Long-term technical debt implications

Format your response with a technical strategy overview, implementation approach, risk assessment, and resource recommendations.

Technical Architect

The Technical Architect persona excels at detailed technical design. Optimize your prompts by:

• Including Technical Specifications: Provide detailed requirements and constraints
• Requesting Design Patterns: Ask for specific architectural approaches
• Specifying Standards Compliance: Mention relevant technical standards
• Including Integration Details: Describe how the solution must interact with existing systems

As a Technical Architect, design a data processing pipeline that meets these requirements:
- Process 10TB of customer transaction data daily
- Maximum latency of 30 minutes for data availability
- Must integrate with our existing PostgreSQL data warehouse
- Should comply with our security standards for PII data

Provide:
- Detailed architectural diagram (describe components)
- Technology selections with rationale
- Data flow specifications
- Integration touchpoints
- Infrastructure requirements
- Error handling approach

Data Analyst

The Data Analyst persona specializes in quantitative analysis. Optimize your prompts by:

• Providing Structured Data Context: Describe datasets, features, and formats
• Specifying Analysis Objectives: Define clear analytical goals
• Including Methodological Preferences: Mention preferred statistical approaches
• Requesting Data-Driven Insights: Ask for patterns, anomalies, or correlations

As a Data analyst, analyze this customer churn dataset to identify key predictors of churn risk.

Dataset contains:
- 50,000 customer records
- 30 features including demographics, usage patterns, and support interactions
- Target variable: churned (yes/no) within 3 months

Perform:
- Exploratory data analysis identifying key patterns
- Feature importance analysis
- Statistical validation of findings (p-values and confidence intervals)
- Recommendations for predictive modeling approach

Format your response with methodology summary, key findings, statistical significance, and actionable insights.

Designer

The Designer persona focuses on user experience and aesthetics. Optimize your prompts by:

• Providing User Context: Describe the target audience and their needs
• Including Experience Goals: Define how users should feel
• Specifying Design Constraints: Mention brand guidelines or technical limitations
• Requesting User-Centered Solutions: Ask for designs that address user pain points

As a Designer, create a conceptual design for our mobile app onboarding experience.

User context:
- Primary users are non-technical professionals ages 35-55
- Main pain points are complexity and time constraints
- Users need to complete account setup, preference selection, and integration setup

Design goals:
- Create an intuitive, step-by-step experience
- Minimize cognitive load and decision fatigue
- Establish trust and professionalism
- Align with our minimalist brand aesthetic (black and white with accent blue #000000)

Provide:
- User flow recommendations
- Key screen concepts (describe visuals)
- Interaction principles
- Success metrics from a user perspective

Sales

The Sales persona specializes in commercial opportunities. Optimize your prompts by:

• Providing Market Context: Describe target markets and competitive landscape
• Including Business Model Elements: Mention pricing, partnership, or revenue details
• Specifying Growth Objectives: Define business development goals
• Requesting Commercial Strategies: Ask for approaches to capture value

As a Sales specialist, develop a go-to-market strategy for our new enterprise productivity tool.

Market context:
- Target market: Financial services firms with 1,000+ employees
- Key competitors: [Competitor A], [Competitor B], and [Competitor C]
- Current pricing model: $50/user/month with annual commitment

Business objectives:
- Achieve 20% market penetration within 18 months
- Establish strategic partnerships with at least 3 major consultancies
- Position as premium solution with high-touch support

Provide:
- Channel strategy recommendations
- Partnership approach with specifics
- Competitive positioning strategies
- Pricing and packaging recommendations
- Key marketing messages aligned with target personas

Expert

The Expert User persona brings domain expertise and practical knowledge. Optimize your prompts by:

• Providing Domain Context: Describe the specific field or industry
• Including Practical Constraints: Mention real-world limitations
• Specifying Experience Level: Define expertise assumptions
• Requesting Applied Knowledge: Ask for practical applications

As an Expert with healthcare administration expertise, evaluate our new patient scheduling system concept.

System context:
- Designed for multi-provider clinics with 10+ practitioners
- Aims to reduce scheduling conflicts and improve resource utilization
- Features include automated appointment reminders, cancellation management, and provider preference matching

From your practical expertise:
- Identify potential implementation challenges in real clinical environments
- Suggest critical features that may be missing
- Evaluate workflow compatibility with typical clinic operations
- Recommend adoption strategies based on your experience

Format your response with a practical assessment, gap analysis, workflow considerations, and implementation recommendations.

User

The User persona specializes in simplifying complex concepts. Optimize your prompts by:

• Defining Knowledge Baseline: Specify assumed knowledge level
• Including Learning Objectives: Define what understanding should be achieved
• Specifying Terminology Constraints: Mention technical terms to avoid or explain
• Requesting Accessible Explanations: Ask for clear, straightforward explanations

As a User with no technical background, explain how our cloud backup system works.

Knowledge context:
- User understands basic computer usage but not networking or cloud concepts
- No prior experience with data backup systems
- Needs to understand enough to use the system confidently

Explain:
- What exactly is being saved when a backup occurs
- Where the data is stored in simple terms
- How security is maintained without technical jargon
- What the user needs to do versus what happens automatically

Format your response with a jargon-free explanation, simple analogies, step-by-step instructions, and answers to common questions.

Text Output

When requesting text output, optimize your prompts by:

• Specifying Format: Indicate document structure (sections, headings, etc.)
• Defining Length: Specify word count or general length guidelines
• Clarifying Tone: Describe desired tone (formal, conversational, technical)
• Indicating Purpose: Explain how the text will be used

Create a customer-facing product explanation document using the TEXT output medium.

Format specifications:
- 4 sections: Overview, Key Features, Use Cases, Getting Started
- 800-1000 words total
- Friendly, approachable tone but still professional
- Include clear headings and short paragraphs for readability
- Optimize for a reader with basic technical knowledge

This document will be used as an introductory resource for new customers exploring our platform.

Media Output

When requesting media output (which will be described rather than generated), optimize your prompts by:

• Specifying Visual Style: Describe aesthetic approach
• Defining Key Elements: List required visual components
• Clarifying Purpose: Explain how the media will be used
• Indicating Audience: Describe who will view the media

Create a concept for a product dashboard using the MEDIA output medium.

Design specifications:
- Minimalist, data-focused design with clean typography
- Primary color scheme: black, white, and blue (#000000)
- Must include: KPI summary section, activity timeline, usage graphs, and alert notifications
- Mobile-responsive layout considerations

The dashboard will be used by marketing managers to track campaign performance. Describe the layout, visual elements, and interaction patterns in detail.

Code Output

When requesting code output, optimize your prompts by:

• Specifying Language/Framework: Indicate programming languages and libraries
• Defining Functionality: Clearly describe what the code should do
• Including Technical Context: Describe systems the code must integrate with
• Setting Code Standards: Mention quality expectations (comments, error handling, etc.)

Create a Python function using the CODE output medium.

Technical specifications:
- Python 3.9+ compatible
- Function to process a CSV of customer transactions and identify potential fraud
- Must use pandas for data manipulation
- Should include proper error handling and logging
- Needs comprehensive docstrings and comments

The function will be integrated into our existing data processing pipeline that handles financial transactions. Include example usage showing how to call the function with sample data.

General Templates

General Information Request

Task: [Specific task description]
Workflow: [Selected workflow pattern]
Personas: [Selected personas]
Output Medium: [Text/Media/Code]

Key Requirements:
- [Requirement 1]
- [Requirement 2]
- [Requirement 3]

Context:
[Relevant background information]

Output Format:
[Desired structure and characteristics]

Comparative Analysis

Task: Compare and analyze [items to compare]
Workflow: [Recommended: Divide & Conquer or Majority Voting]
Personas: [Selected personas]

Comparison Criteria:
- [Criterion 1]
- [Criterion 2]
- [Criterion 3]

For Each Item, Analyze:
- Strengths and weaknesses
- Unique characteristics
- Best application scenarios

Output Format:
- Comparative summary
- Detailed analysis of each item
- Recommendations based on specific use cases

Problem-Solution Framework

Task: Develop a solution for [problem statement]
Workflow: [Recommended: Sequential Refinement or Creator-Critic]
Personas: [Selected personas]

Problem Context:
[Detailed description of the problem]

Constraints:
- [Constraint 1]
- [Constraint 2]
- [Constraint 3]

Solution Requirements:
- [Requirement 1]
- [Requirement 2]
- [Requirement 3]

Output Format:
- Problem analysis
- Proposed solution(s)
- Implementation considerations
- Expected outcomes

Task-Specific Templates

Product Documentation Template

Task: Create comprehensive product documentation
Workflow: Sequential Refinement

Stage 1 (Technical Architect):
- Develop technical specifications
- Document system architecture
- Define feature functionality in technical terms

Stage 2 (Smart User):
- Transform technical specifications into practical explanations
- Add usage scenarios and examples
- Identify common questions and provide answers

Stage 3 (Designer):
- Improve document structure and readability
- Add conceptual illustrations (described)
- Enhance navigation and information hierarchy

Output Format:
- Complete product documentation with:
  * Product overview
  * Feature specifications
  * Setup instructions
  * Usage guidelines
  * Troubleshooting section
  * FAQ

Marketing Strategy Template

Task: Develop a comprehensive marketing strategy
Workflow: Divide & Conquer

Subtask 1 (Data analyst):
- Market analysis and target audience identification
- Competitive positioning analysis
- Performance metrics and measurement framework

Subtask 2 (Sales):
- Channel strategy and partnership opportunities
- Pricing and packaging recommendations
- Revenue projections and budget allocation

Subtask 3 (Designer):
- Brand messaging and value proposition
- Customer journey and touchpoint strategy
- Creative direction and campaign concepts

Integration Guidelines:
- Ensure consistent positioning across all components
- Align metrics with channel strategies
- Create cohesive customer experience across touchpoints

Output Format:
- Complete marketing strategy including:
  * Executive summary
  * Market analysis
  * Target audience profiles
  * Positioning strategy
  * Channel recommendations
  * Budget and timeline
  * Success metrics

Technical Evaluation Template

Task: Evaluate technical solutions for [specific need]
Workflow: Majority Voting

Decision Question:
Which technical solution best meets our requirements for [specific need]?

Options to Evaluate:
- Solution A: [description]
- Solution B: [description]
- Solution C: [description]

Evaluation Criteria:
- Technical capabilities and limitations
- Implementation complexity
- Scalability and performance
- Maintenance requirements
- Cost and resource implications
- Ecosystem and support considerations

Agent Instructions (for CTO, Technical Architect, and Smart User):
- Evaluate each solution against all criteria
- Rate each solution on a 1-10 scale for each criterion
- Provide specific reasoning for ratings
- Make a final recommendation with justification

Output Format:
- Comparison matrix with ratings
- Detailed analysis of each solution
- Consensus recommendation with rationale
- Implementation considerations for recommended solution

Precision & Clarity

To ensure precise and clear prompts:

1. Use Specific Verbs: Replace "analyze" with "identify key trends" or "compare quarterly performance"
2. Quantify When Possible: Include specific numbers for length, time periods, or quantities
3. Define Ambiguous Terms: Clarify what you mean by terms like "best," "optimize," or "improve"
4. Use Bulleted Lists: Break down complex requirements into discrete items
5. Provide Examples: Include illustrations of desired outputs when applicable

Analyze our marketing data and give recommendations.

Analyze our Q1-Q2 2025 marketing campaign data to:
- Identify the top 3 performing channels by conversion rate
- Compare cost-per-acquisition across different customer segments
- Recommend 2-3 specific optimizations for our Q3 strategy based on these insights

Format your analysis with specific metrics, visualizations (described), and actionable recommendations.

Contextualization

To provide effective context in your prompts:

1. Include Relevant Background: Add information about why the task matters
2. Define Audience: Specify who will use the output
3. Explain Constraints: Mention limitations that should be considered
4. Connect to Larger Goals: Link the task to broader objectives
5. Provide Reference Points: Include examples, benchmarks, or standards

Create a feature requirements document for our new mobile app notification system.

Create a feature requirements document for our new mobile app notification system.

Context:
- Our users have reported 40% notification fatigue in recent surveys
- Competitors have introduced customizable notification settings
- Our technical stack uses Firebase for notification delivery
- This feature is part of our Q3 initiative to improve user engagement

The document will be used by our development team to implement the feature and by our QA team to validate functionality.

Iteration & Refinement

To effectively iterate on your prompts:

1. Start Simple: Begin with a basic version of your prompt
2. Evaluate Outputs: Analyze results against your expectations
3. Identify Gaps: Determine what's missing or incorrect
4. Add Specificity: Enhance the prompt with additional details
5. Test Variations: Try different approaches to the same task

Create a weekly social media content plan.

Create a B2B weekly social media content plan for LinkedIn and Twitter.

Create a weekly B2B social media content plan for LinkedIn and Twitter that highlights our AI Workflow Grid product.

Include:
- 3 LinkedIn posts focused on thought leadership
- 5 Twitter posts with industry insights
- Recommended posting schedule
- Content themes aligned with our Q2 marketing focus on "Enterprise AI Efficiency"
- Suggested formats (text, image descriptions, etc.)

Our target audience is enterprise IT decision-makers in financial services and healthcare industries.

Error Recovery

When outputs don't meet expectations:

1. Analyze the Issue: Identify specific problems with the output
2. Provide Feedback: Explain what wasn't right about the previous result
3. Clarify Requirements: Add details that were missing from the original prompt
4. Add Constraints: Include what to avoid or exclude
5. Use Contrast: Explain what you want instead of what you received

Write a case study about implementing our AI Workflow Grid.

The previous case study was too generic. Instead, write a detailed case study about XYZ Financial implementing our AI Workflow Grid.

Specifically include:
- Their challenge: High volume document processing with compliance requirements
- Implementation details: Which specific workflows and personas they utilized
- Quantifiable results: 65% reduction in processing time, 40% cost savings
- Quotes from their CTO (fictional but realistic)
- Do NOT include generic benefits without specific metrics
- Do NOT use placeholder text like "XX% improvement" - use the specific numbers provided

The case study should be concrete, specific to financial services compliance, and focused on measurable outcomes.

Compliance & Security

For regulated industries or sensitive information:

1. Avoid PII in Prompts: Never include personally identifiable information
2. Use Abstraction: Replace specific details with general placeholders
3. Apply Minimalism: Include only necessary information
4. Add Compliance Instructions: Explicitly mention regulatory requirements
5. Include Disclaimer Requirements: Specify when disclaimers are needed

Task: Create a template for financial product descriptions
Workflow: Sequential Refinement
Personas: Expert, Sales, User

Create a template that financial advisors can use to describe investment products.

Compliance requirements:
- Must include standard risk disclaimer language: "Investment involves risk. Past performance is not indicative of future results."
- Avoid absolute claims about performance or returns
- Include sections for fee disclosure
- Use clear language avoiding industry jargon per regulatory guidelines
- Don't include specific investment examples at this stage

Create the template with placeholder sections that advisors can customize while maintaining compliance.

Standardizing Prompting Practices

For enterprise-wide consistency:

1. Create Prompt Libraries: Develop organization-specific templates
2. Establish Style Guides: Define standard formats and approaches
3. Implement Review Processes: Set up quality checks for important prompts
4. Develop Training: Create educational materials for teams
5. Document Best Practices: Maintain living documentation of effective techniques

[COMPANY NAME]

Purpose: [Brief description of prompt objective]
Intended Output: [Description of expected deliverable]
Workflow Pattern: [Selected workflow]
Assigned Personas: [Selected personas]

Task Specifications:
- [Detailed requirement 1]
- [Detailed requirement 2]
- [Detailed requirement 3]

Company Context:
- [Relevant company information]
- [Brand guidelines summary]
- [Relevant policies]

Output Requirements:
- [Format specifications]
- [Quality criteria]
- [Compliance requirements]

Approvals: [Required approvers]
Last Updated: [Date]
Version: [Number]

Common Issues & Solutions

Debugging Prompt Performance

When prompts don't produce expected results, use this diagnostic process:

1. Isolate Variables: Change one element at a time to identify issues
   • Test different workflows with same prompt
   • Test different personas with same workflow and prompt
   • Test variations of prompt language while keeping other elements constant
2. Simplify and Rebuild: Start with minimal prompts and add complexity
   • Begin with core instruction only
   • Add context elements one by one
   • Track which additions improve or degrade performance
3. Target Specific Issues: Address known problems directly
   • For outputs that are too long: "Limit response to 500 words maximum."
   • For technical inaccuracies: "Verify all technical statements against [standard]."
   • For tone problems: "Maintain a professional, objective tone throughout."
4. Use Explicit Feedback: Tell the system what went wrong

   The previous response was too technical for our audience. Please revise with these changes:
   - Use simpler language equivalent to high school reading level
   - Replace technical jargon with plain language explanations
   - Add brief explanations for necessary technical terms
   - Include more real-world examples to illustrate concepts