A multi-agent AI system that simulates Accenture's Application Delivery Method (ADM) lifecycle. Built with CrewAI and powered by Nebius Token Factory inference.
This project deploys 11 AI agents — each representing a distinct Accenture delivery role — to collaboratively execute a full enterprise engagement lifecycle. A Managing Director agent orchestrates the crew hierarchically, delegating work across strategy, consulting, architecture, development, testing, and deployment phases.
The system accepts a client name and engagement type as input and produces structured deliverables across 12 ADM phases: from initial engagement strategy through solution architecture, sprint execution, UAT coordination, production deployment, and executive closure.
A Chainlit web UI enables real-time client interaction: clients describe their project, watch the crew work through each phase with streaming output, and answer clarifying questions from the consulting team directly in the chat.
Key capabilities:
- Hierarchical agent orchestration with delegation across 11 specialist roles
- 12-phase ADM lifecycle execution with task dependencies and context passing
- Interactive client-facing chat UI with human-in-the-loop questioning
- Web search (Tavily) and site scraping tools for real-world research
- Streaming output with per-phase progress visualization
- Final executive synthesis report generated as a deliverable
Tech stack: CrewAI 1.13 | DeepSeek V3.2 via Nebius Token Factory | Tavily Search | Chainlit | Python 3.11+
┌─────────────────────────────────────────────────────────────┐
│ Chainlit Web UI │
│ Client chat + streaming output │
└───────────────────────┬─────────────────────────────────────┘
│
▼
┌─────────────────────────────────────────────────────────────┐
│ Managing Director (Manager Agent) │
│ Orchestrates all tasks hierarchically │
│ DeepSeek V3.2 │
└───────────────────────┬─────────────────────────────────────┘
│ delegates to
┌───────────────┼───────────────┐
▼ ▼ ▼
┌──────────────┐ ┌──────────────┐ ┌──────────────┐
│ Strategy & │ │ Technology │ │ Delivery & │
│ Consulting │ │ Track │ │ Operations │
│ │ │ │ │ │
│ • Assoc Dir │ │ • Solution │ │ • Sr Delivery│
│ • Engagement │ │ Architect │ │ Lead │
│ Manager │ │ • Technology │ │ • Program │
│ • Mgmt │ │ Architect │ │ Mgmt Lead │
│ Consultant │ │ • Tech │ │ • Digital PM │
│ • Strategy │ │ Delivery │ │ │
│ Analyst │ │ Lead │ │ │
└──────────────┘ └──────────────┘ └──────────────┘
│ │ │
▼ ▼ ▼
┌─────────┐ ┌─────────┐ ┌─────────┐
│ Tavily │ │ Tavily │ │ Tavily │
│ Search │ │ Search │ │ Search │
│ Scrape │ │ Scrape │ │ Scrape │
│ Ask │ │ │ │ │
│ Client │ │ │ │ │
└─────────┘ └─────────┘ └─────────┘
The crew executes 12 tasks in sequence, each assigned to a specialist agent. Tasks receive context from their predecessors, building a coherent engagement narrative.
| # | Phase | Agent | Tools |
|---|---|---|---|
| 1 | Engagement Strategy & Delivery Planning | Associate Director | Search, Scrape, Ask Client |
| 2 | Strategic Engagement Approval | Senior Delivery Lead | Search, Scrape, Ask Client |
| 3 | Business Requirements Analysis | Strategy & Consulting Analyst | Search, Scrape, Ask Client |
| 4 | Backlog Definition & Scope Planning | Management Consultant | Search, Scrape, Ask Client |
| 5 | Solution Architecture Design | Solution Architect | Search, Scrape |
| 6 | Technical Architecture Review | Technology Architect | Search, Scrape |
| 7 | Solution Development & Implementation | Technology Delivery Lead | Search, Scrape |
| 8 | Sprint Execution & Project Tracking | Digital Project Manager | Search, Scrape |
| 9 | Integrated Testing & Solution Validation | Engagement Manager | Search, Scrape, Ask Client |
| 10 | User Acceptance Testing Coordination | Engagement Manager | Search, Scrape, Ask Client |
| 11 | Production Deployment & Cutover | Program Management Lead | Search, Scrape |
| 12 | Service Introduction & Engagement Closure | Associate Director | Search, Scrape, Ask Client |
The final task receives context from all 11 preceding phases to produce a comprehensive executive synthesis report.
├── app.py # Chainlit web UI entry point
├── pyproject.toml # Project config & dependencies
├── .env # API keys (not committed)
├── .chainlit/config.toml # Chainlit UI configuration
└── src/accenture_adm_hierarchical_delivery_crew/
├── crew.py # Crew orchestration & agent definitions
├── main.py # CLI entry point
├── config/
│ ├── agents.yaml # Agent roles, goals, backstories
│ └── tasks.yaml # Task descriptions, outputs, dependencies
└── tools/
├── __init__.py
└── ask_human.py # Human-in-the-loop tool (Chainlit integration)
- Python 3.11+ (3.13 recommended)
- uv for dependency management
- A Nebius Token Factory API key
- A Tavily API key
# Clone the repo
git clone https://github.com/colygon/accenture-adm-crew.git
cd accenture-adm-crew
# Create venv and install dependencies
uv venv --python 3.13
uv sync
uv pip install chainlit tavily-pythonCreate a .env file in the project root:
NEBIUS_API_KEY=your_nebius_token_factory_key
TAVILY_API_KEY=your_tavily_api_key.venv/bin/chainlit run app.pyOpens at http://localhost:8000. Enter your company name and engagement type, then interact with the consulting team as they work through each ADM phase.
# Interactive prompts for client name and engagement type
.venv/bin/python -m accenture_adm_hierarchical_delivery_crew.main run
# Or via CrewAI CLI
crewai run# Train the crew over N iterations
.venv/bin/python -m accenture_adm_hierarchical_delivery_crew.main train <n_iterations> <output_file>
# Replay from a specific task
.venv/bin/python -m accenture_adm_hierarchical_delivery_crew.main replay <task_id>All agents use DeepSeek V3.2 via Nebius Token Factory ($0.30/$0.45 per 1M tokens). The model is configured in crew.py:
default_llm = LLM(
model="openai/deepseek-ai/DeepSeek-V3.2",
base_url="https://api.tokenfactory.nebius.com/v1",
api_key=NEBIUS_API_KEY,
)To use a different model, swap the model ID. Available options include:
| Model | Type | Cost (in/out per 1M) |
|---|---|---|
deepseek-ai/DeepSeek-V3.2 |
Chat | $0.30 / $0.45 |
Qwen/Qwen3.5-397B-A17B |
Chat | $0.60 / $3.60 |
zai-org/GLM-5 |
Chat | $1.00 / $3.20 |
deepseek-ai/DeepSeek-R1-0528 |
Reasoning | $0.80 / $2.40 |
NousResearch/Hermes-4-405B |
Reasoning | $1.00 / $3.00 |
All model IDs must be prefixed with openai/ for litellm routing (e.g., openai/deepseek-ai/DeepSeek-V3.2).
Each ADM phase writes its output to deliverables/ as a numbered markdown file:
| File | Content |
|---|---|
01_engagement_strategy.md |
Engagement strategy & delivery planning |
02_strategic_approval.md |
Go/no-go approval decision |
03_business_requirements.md |
Functional & non-functional requirements |
04_backlog_scope.md |
Prioritized backlog with acceptance criteria |
05_solution_architecture.md |
End-to-end architecture design |
06_technical_architecture_review.md |
Architecture validation & risk assessment |
07_development_implementation.md |
Solution increment & build artifacts |
08_sprint_execution.md |
Sprint metrics & progress tracking |
09_testing_validation.md |
Test coverage & release readiness |
10_uat_coordination.md |
User acceptance testing results |
11_deployment_cutover.md |
Go-live execution & stability metrics |
12_engagement_closure.md |
Executive synthesis & strategic recommendations |
In the Chainlit UI, deliverables are viewable and downloadable after the engagement completes. Type a number to view a specific deliverable or all to browse everything.
Private repository. All rights reserved.