Consulting Agent Crew

A DeepAgent-grade consulting system powered by crewAI, designed to deliver MBB-quality strategic consulting through hypothesis-driven analysis, rigorous quality gates, and evidence-backed recommendations.

This system orchestrates 8 specialized consulting agents through a 5-phase workflow to produce enterprise-grade strategic consulting deliverables. Each agent operates with access to company knowledge bases (RAG) and web search capabilities, ensuring recommendations are grounded in both internal context and external market realities.

Features

• Hypothesis-Driven Workflow: All analysis mapped to testable hypotheses with explicit evidence tagging
• 5-Phase Structure: Sequential phases with explicit quality gates that can loop back if standards aren't met
• Evidence Tagging: Every finding tagged as supports/weakens/inconclusive relative to hypotheses
• Quality Control Gates: Explicit gates prevent weak recommendations from proceeding
• RAG + Web Search: Company knowledge base search with automatic fallback to web search for current market data
• Pressure-Testing: Recommendations tested against hostile board scenarios, assumption failures, and execution risks
• 8 Specialized Agents: Each agent has a specific role (Principal/Partner-level) matching MBB consulting structures

Installation

Ensure you have Python >=3.10 <3.14 installed on your system. This project uses UV for dependency management.

First, install uv:

pip install uv

Next, navigate to your project directory and install the dependencies:

crewai install

Configuration

Add your API keys into the .env file:

• OPENAI_API_KEY - Required for LLM functionality
• SERPER_API_KEY - Required for web search (get your key from serper.dev)

Customize the consulting engagement:

• Modify config/agents.yaml to adjust agent roles, goals, and backstories
• Modify config/tasks.yaml to customize phase tasks and requirements
• Modify crew.py to add custom tools or modify agent configurations
• Modify main.py to change the default client inputs (client_company, strategic_objective, target_market)
• Add company knowledge documents to knowledge/ directory (company_overview.md, company_strategy.md, company_capabilities.md, etc.)

Running the Project

To run the consulting crew with default inputs (TechCorp Global example), execute from the root folder:

crewai run

Or run directly with Python:

python main.py

This initializes the Consulting Agent Crew, assembling all 8 agents and executing tasks sequentially through the 5-phase workflow. Each phase produces markdown deliverables that are saved to the project root.

Project Structure

.
├── config/
│   ├── agents.yaml          # 8 agent definitions (roles, goals, backstories)
│   └── tasks.yaml           # 5-phase task definitions with quality gates
├── knowledge/               # Company knowledge base (RAG)
│   ├── company_overview.md  # Company profile, mission, values, structure
│   ├── company_strategy.md  # Strategic priorities and initiatives
│   ├── company_capabilities.md  # Organizational capabilities
│   └── user_preference.txt  # User preferences and context
├── tools/                    # Custom tools
│   └── custom_tool.py       # Additional custom tools (if needed)
├── crew.py                  # Crew class with 8 agents and 11 tasks
├── main.py                  # Entry point with default client inputs
└── pyproject.toml           # Project configuration and dependencies

Understanding the Crew

The 8 Agents

The Consulting Agent Crew consists of 8 specialized agents, each with Principal or Partner-level expertise:

1. Project Orchestration Agent (Engagement Manager) - Coordinates workflow, manages dependencies, tracks deliverables
2. Research Intelligence Agent (Principal) - Industry reports, competitive analysis, market intelligence, PESTEL analysis
3. Data Analysis Agent (Principal) - Quantitative analysis, financial modeling, statistical analysis, forecasting
4. Strategic Framework Agent (Partner) - Applies Porter's Five Forces, SWOT, Value Chain, strategic frameworks
5. Interview Synthesis Agent (Principal) - Stakeholder interviews, qualitative synthesis, organizational analysis
6. Recommendation Engine Agent (Partner) - Generates strategic recommendations, implementation roadmaps, ROI estimates
7. Document Production Agent (Principal) - Creates executive summaries, reports, slide decks, tailored messaging
8. Quality Control Agent (Partner) - Enforces quality gates, challenges assumptions, validates evidence quality

All agents have access to:

• Company Knowledge Base Search (DirectorySearchTool) - Searches internal company documents
• File Reader (FileReadTool) - Reads specific company documents in detail
• Web Search (SerperDevTool) - Searches web for current market data when company docs lack context

The 5-Phase Workflow

The crew operates through a rigorous 5-phase consulting process:

Phase 0: Engagement Framing

• Problem statement (root causes, not symptoms)
• Success metrics
• 2-4 testable hypotheses (not solutions)
• Explicit unknowns
• Initial problem structuring (hypothesis/issue tree)
• Agent: Project Orchestration Agent
• Output: phase_0_engagement_framing.md

Phase 1: Hypothesis-Driven Discovery (Parallel Work) Three parallel tasks that test hypotheses:

• Research Intelligence: Industry analysis, competitive landscape, market trends, PESTEL
• Data Analysis: Financial modeling, quantitative analysis, scenario analysis, benchmarking
• Interview Synthesis: Stakeholder interviews, qualitative insights, organizational assessment
• Agents: Research Intelligence Agent, Data Analysis Agent, Interview Synthesis Agent
• Outputs: phase_1_research_intelligence_report.md, phase_1_data_analysis_report.md, phase_1_interview_synthesis_report.md
• Critical: Every finding tagged as SUPPORTS/WEAKENS/INCONCLUSIVE for each hypothesis

Phase 2: Synthesis & Pushback (Quality Gate)

• Synthesize all Phase 1 evidence
• Hypothesis validation matrix
• Refined problem framing
• Shortlist of 2-4 strategic options
• Explicit trade-offs
• Framework analysis (Porter's Five Forces, SWOT, Value Chain)
• Quality Gate Decision: PROCEED / LOOP BACK / REFRAME
• Agents: Strategic Framework Agent, Quality Control Agent
• Outputs: phase_2_synthesis_pushback_report.md, phase_2_quality_gate_decision.md

Phase 3: Strategy Design & Option Testing

• Detailed option analysis (Build/Buy/Partner/Hybrid)
• Financial impact (3-5 year projections)
• Execution complexity assessment
• Risk profiles and mitigation
• Organizational readiness evaluation
• Financial modeling and scenario analysis
• Agents: Recommendation Engine Agent, Data Analysis Agent
• Outputs: phase_3_strategy_design_report.md, phase_3_data_modeling_report.md

Phase 4: Recommendation Pressure-Test (Critical Gate)

• Pressure-test against hostile board scenarios
• Test assumption failures
• Evaluate execution risks
• Assess competitive response
• Validate organizational reality
• Gate Decision: PROCEED / REFINE
• Agent: Quality Control Agent
• Output: phase_4_pressure_test_report.md

Phase 5: Narrative & Decision Packaging

• Executive Summary (decision-first approach)
• Detailed Strategic Report (hypothesis-led narrative)
• Executive Presentation Deck outline
• Tailored messaging (C-suite vs operational)
• Comprehensive appendices
• Agent: Document Production Agent
• Output: phase_5_consulting_deliverables.md

Quality Gates

The system includes explicit quality gates that can loop back to previous phases:

• Phase 2 Gate: If evidence is weak → loop back to Phase 1; if hypotheses wrong → reframe in Phase 0
• Phase 4 Gate: If recommendations fail pressure-test → send back to Phase 3 for refinement

This ensures only defensible, evidence-backed recommendations reach the client.