Self-improving AI agent infrastructure that gets smarter with every use
Enterprise-scale modular architecture for strategic consistency automation
🧠 Example Map Learning System
- Accumulated Wisdom Architecture: Agents read ALL evaluation files (0 to 100+) for pattern recognition
- Dual Learning Model: High scores become benchmarks, low scores teach failure avoidance
- Quality Filtering: 8.0+ threshold creates reference patterns for replication
- No Evaluation Required: System works perfectly with zero evaluations - each evaluation just makes it exponentially better
- Model-Training Ready: Builds a valuable, structured dataset (5_labeled_json/) perfect for fine-tuning custom models.
🏗️ Modular Plugin Architecture (Salesforce Model)
Core System (Works Independently)
├── Agent 1: Message House
├── Agent 2: Brand Personas
├── Agent 3: Customer Reality Analysis
├── Agent 4: Gap Analysis
├── Agent 5: Keywords Engine (Quality Gate)
├── Agent 7: Testimonial Generation
├── Agent 8: Social Media Content
└── Agent 9: Website Copy with Psychology Logic
Optional Enhancements (Premium Features)
├── Agent 6: Stakeholder Strategy Decks
├── Agent 6.5: Platform Intelligence Plugin
└── Agent 10: Consistency Validation
🔌 Template Expansion Framework
- Agent Development Template: Build new agents using proven patterns
- Industry-Specific Optimization: Auto-generated system prompts per business domain
- Script Pattern Library: Copy-paste code templates for rapid deployment
- Integration Checklist: Systematic pipeline integration guide
The Expensive Reality: Marketing teams face chaotic disconnect between strategy, customer reality, and content execution. Teams burn countless hours trying to align campaigns with core messaging, only to find efforts are inconsistent, out of touch with real customers, and inefficient to update—leaking budget and diluting brand with every initiative.
Our Technical Solution: Strategic consistency engine with enterprise-scale automation.
This engine doesn't just generate text; it delivers a complete, cohesive, and production-ready Go-To-Market portfolio from a single strategic input. Key assets include:
- A foundational Message House & core brand documents.
- Authentic, data-driven Customer and Brand Personas.
- A stakeholder-ready Strategic Gap Analysis report.
- A comprehensive SEO-optimized Keywords Bank.
- A library of compelling marketing Testimonials.
- A multi-faceted Social Media Content Strategy.
- Psychology-driven, long-form Website Copy.
Agent 0a: Project Configurator
- Adaptive Prompt Engineering: Auto-generates unique system prompts for all downstream agents, tailored to the project's specific industry and product type for higher-quality, context-aware output.
- Multi-Project Isolation: Workspace creation with zero cross-contamination
- 8-Agent Configuration: Simultaneous setup across entire core pipeline
Agent 0b: Pipeline Orchestrator
- Two-Phase Execution: Foundation building → Quality gate → Content generation
- State Persistence: Resume from any interruption point with JSON tracking
- Intelligent Mode Detection: Validation vs Brand Mode based on available files
- Dependency Management: Smart file copying and prerequisite validation
Agent 1: Message House Engine
- Input: PRE-SYSTEM Human Expert Input: 12 strategic questions that distill core business conviction and product psychology.
- Output: Strategic messaging foundation for entire pipeline
- Learning: Example Map system with 8.5+ quality targeting
- Integration: Feeds Agents 2, 5, 8, 9 directly
Agent 2: Brand Persona Generator
- Innovation: Transforms official brand message into ideal user personas
- Learning: Example Map system for persona authenticity patterns
- Fallback Role: Feeds testimonials and social media in Brand Mode when customer data unavailable
Agent 3: Customer Reality Analyzer
- Innovation: Extracts authentic personas from real customer reviews
- Strategic Decision: No iterative learning to preserve fresh analytical reasoning
- Data Source: CSV files with customer review content and ratings
Agent 4: Strategic Gap Analysis
- Framework: 4-phase systematic methodology with quantitative evidence
- Innovation: Identifies invisible gaps between brand intention and customer reality
- Decision-Ready Insights: Transforms qualitative gaps into a quantitative, prioritized action plan, equipping stakeholders to make data-driven decisions on strategy.
- Output: Strategic discovery insights for positioning refinement
Agent 5: Keywords Bank (Two-Phase Architecture)
- Phase 1: Strategic foundation generation with human evaluation requirement
- Quality Gate: ≥7.0 score threshold prevents strategy drift
- Phase 2: Massive expansion engine (150+ keywords across 6 vectors)
- Innovation: Solves Internal→External asset misalignment problem
Agent 7: Testimonial Engine
- Strategic Blending Framework: Weaves brand messaging into authentic customer voice
- Multi-Format Output: Short, medium, long-form testimonials with diverse profiles
- Learning: Example Map system for authenticity pattern recognition
Agent 8: Social Media Content Generator
- Character Innovation: "Emma" - dedicated brand ambassador with consistent voice
- 4-Category Framework: Real Experience, Community, Product Feature, Brand Info
- Content Recipe Matrix: 12 proven patterns (3 recipes × 4 categories)
- Platform Intelligence: Optional Agent 6.5 enhancement for Twitter-specific optimization
Agent 9: Website Copy with Strategic Logic
- Core Innovation: Generates compelling psychology reasoning BEFORE content
- Two-Phase Process: 1) Sell the strategic logic, 2) Provide content modules
- Strategic Decision: No iterative learning to maintain creative freshness
- 7 Content Modules: Complete homepage architecture with conversion psychology
{agent_name}/
├── 1_input/{project_name}/ ← Project-specific inputs
├── 2_system_assets/{project_name}/ ← Industry-optimized prompts
├── 3_unlabeled/{project_name}/ ← Project-specific outputs
├── 4_labeled_md/ ← Shared quality assets
└── 5_labeled_json/ ← Shared learning data
- Project Detection: Auto-discovery across all agents
- State Tracking: Complete execution history with resume capability
- Error Recovery: Graceful failure handling with retry mechanisms
- Quality Assurance: Multi-stage validation preventing cascade failures
- Parallel Execution: Independent agents run simultaneously where possible
- Dependency Optimization: Smart file detection and propagation
- Resource Management: Minimal API calls with maximum output quality
- Enterprise Security: Project isolation with no data leakage
- API: Claude 3.5 Sonnet via Anthropic API
- Language: Python 3.x with minimal dependencies
- Architecture: Microservices approach with standalone agents
- Data: JSON-based state management with UTF-8 encoding
- Platform: Cross-platform compatibility (Windows/Mac/Linux)
- Template System: Standardized agent creation patterns
- Integration Standards: Pipeline compatibility requirements
- Evaluation Interface: GUI-based scoring and tagging system
- Learning Infrastructure: Example Map architecture for continuous improvement
- Human-in-the-Loop: Strategic evaluation points where it matters most
- Automated Validation: Dependency checking and file verification
- Progressive Enhancement: System works perfectly without evaluation, gets better with it
- Business Logic Preservation: Strategic agents maintain creative freshness
# 1. Clone repository
git clone https://github.com/Ace-Hsu/AI-Go-To-Market-Engine.git
# 2. Configure API
cp config.template.json config.json
# Add your Anthropic API key
# 3. Run any agent
cd agents/message_house_agent
python scripts/generate_simple.py# 1. Project Configuration
cd agents/agent_0a_configurator
python scripts/generate_simple.py
# Interactive setup: product type, industry focus (target market)
# 2. Pipeline Execution Phase 1
cd ../agent_0b_orchestrator
python scripts/run_pipeline_phase1.py
# Foundation building: Agents 1-4, plus Agent 5 Keywords Phase 1
# 3. Quality Gate
cd ../keywords_bank_agent
python scripts/evaluate_phase1.py
# Human evaluation: ≥7.0 score required
# 4. Pipeline Execution Phase 2
cd ../agent_0b_orchestrator
python scripts/run_pipeline_phase2.py
# Content generation: Agent 5 Keywords Expansion, then Agents 7,8,9- Master Brand Consistency Across Portfolios: Ensure a single, unified strategic voice across a portfolio of 10+ products, eliminating brand dilution and inconsistent messaging at scale.
- De-risk Strategic Drift: Lock in core messaging with mandatory human-in-the-loop quality gates before any content is generated, guaranteeing alignment.
- Automate High-Quality Asset Production: Move from manual, multi-week content cycles to a fully automated pipeline that delivers a complete asset kit in under an hour.
- Accelerate Time-to-Market: Onboard new products or campaigns with a system that instantly generates all foundational marketing materials from a single strategic input.
- Deliver Superior Client Results: Utilize secure project isolation to manage multiple clients, ensuring high-quality, strategically consistent outputs for each one.
- Onboard New Clients in Hours, Not Weeks: Leverage the template framework to rapidly configure and deploy the entire content engine for new clients, dramatically reducing setup time.
- Offer a Unique "Learning System" Advantage: Differentiate your agency by offering a content engine that gets progressively smarter and more attuned to your client's brand with every feedback cycle.
- Boost Agency Efficiency and Profitability: Automate the most time-consuming aspects of content strategy and creation, freeing up your team to focus on higher-value client services.
- Accelerate Your Own AI Agent Development: Use this repository as a complete, production-grade blueprint for building complex, multi-agent workflows with persistent state.
- Implement Advanced Human-in-the-Loop Learning: Adopt the "Example Map" architecture as a proven pattern for creating systems that continuously improve based on structured human feedback.
- Avoid Monolithic Designs: Build resilient and scalable applications by studying our modular plugin architecture.
- Example Map Learning: Agents learn from complete evaluation history, not single examples
- Strategic Logic Generation: Fresh psychology reasoning for each brand context
- Two-Phase Quality Control: Prevents strategy drift at enterprise scale
- Internal→External Asset Transfer: Solves messaging consistency problem systematically
- Learning vs Freshness: Content agents learn patterns, strategic agents maintain creativity
- Quality Gates: Human evaluation where it matters most, automation where it scales
- Plugin Model: Core functionality works independently, enhancements add value
- Enterprise First: Multi-project architecture from day one, not retrofitted
- Uses 1-5: Good baseline quality with initial pattern recognition
- Uses 5-10: Strong pattern recognition emerges across agent types
- Uses 10+: Expert-level insights with deep strategic sophistication
- Uses 50+: Enterprise-level intelligence with domain expertise
- Target Scores: 8.5+ for business-ready assets across all agents
- Evaluation Criteria: Strategic depth, customer psychology, business impact focus
- Success Metrics: Complete asset portfolios with strategic alignment
- Business Validation: Direct usability in marketing campaigns and operations
This engine's first mission is to serve as a single source of truth that solves the core problem of strategic disconnect. But solving for consistency is just the beginning. The long-term vision is to evolve this into a self-optimizing growth platform. The architecture is designed not just to execute strategy, but to learn from real-world results and make the entire GTM process smarter over time.
Core Capabilities:
- Platform Content Agents: Tailored content generation with platform-specific intelligence for LinkedIn, Email, and Instagram.
- Demo Video Agent: Transforms core GTM assets into compelling video narratives, complete with auto-generated scripts and visuals.
- Product Dictionary Agent: RAG-powered access to all GTM assets for instant, product-specific answers.
- Public Showcase: Launch a public /examples directory with curated user feedback and learning files (5_labeled_json/) to transparently demonstrate system value.
Advanced Strategic Intelligence:
- PersonaThink Customer Research System: A new core engine that executes end-to-end customer research—from focus group selection to survey simulation—delivering the insight report that powers Agent 3 (Customer Reality) and Agent 4 (Gap Analysis).
The Self-Optimizing Loop:
- Performance Feedback Integration: Measures asset performance via a connected analytics dashboard to identify and scale what truly works.
A complete, production-ready AI infrastructure that bridges business strategy with technical execution at enterprise scale.