🚀 Agentic Corporate Simulation: A Cognitive Multi-Agent Framework

In a modern corporate landscape, decision-making often requires the intersection of various expertise—Strategic, Human Resources, and Operational. This project builds an Agentic AI Ecosystem that simulates a professional board of directors.

Instead of a simple chatbot, this system implements a Cognitive Architecture where specialized AI Agents (CEO, CHRO, Manager) collaborate to solve business problems. By treating agents as modular tools and implementing a sophisticated memory management cycle, the framework ensures high-quality reasoning while maintaining efficiency in production environments.

This repository features a sophisticated Agentic AI System designed to simulate a corporate decision-making environment. It leverages a Supervisor-Worker architecture, served via FastAPI, and is fully containerized using Docker.

🔄 1. System Workflow: The Lifecycle of a Request

The system utilizes a LangGraph state machine to ensure logical consistency and context efficiency through a predefined execution path:

🔹 Entry Point (Summarizer Node)

Every user input is first processed by the Summarizer.
If the thread history exceeds 10 messages:
- Older interactions are archived into a “Progressive Summary”.
- Only the 3 most recent messages are retained as raw context.
This ensures the LLM receives a condensed yet relevant context window, preventing context overflow and maintaining focus.

🔹 Routing (Orchestrator / Supervisor Node)

The Orchestrator analyzes the user intent alongside the current summary.
Acts as a router using LangChain Tool Calling to trigger specialized sub-agents:
- call_ceo → Strategy
- call_chro → HR

🔹 Execution (Worker Nodes)

The selected Worker (CEO, CHRO, or Manager) receives refined context.
Generates responses based on:
- Persona
- Business Goal (defined in YAML configurations)

🔹 Persistence

State is automatically saved to PostgreSQL via an asynchronous checkpointer.
Enables stateful, long-running conversations across sessions.

🧠 2. Intelligent Memory Management

🔹 Short-Term Memory

Managed via LangGraph State
Tracks current conversation flow

🔹 Progressive Memory

The Summarizer Agent compresses historical data
Maintains long-term context while optimizing token usage

🔹 Stateful Persistence

Powered by AsyncPostgresSaver
Allows perfect restoration of any conversation thread

🛠️ 3. Tech Stack & Infrastructure

Framework: LangChain (Prompt Templates, LLM interaction, Output Parsing)
Orchestration: LangGraph (State Machine & cyclic workflows)
Infrastructure: Docker & Docker Compose
Backend: FastAPI (Async production-grade API)
LLM Gateway: OpenRouter (gpt-oss-120b:free)
Database: PostgreSQL (Async thread-state checkpointer)

🔮 4. Future Roadmap: Towards a Cognitive Engine

🔹 Modular Agent Toolsets

Role-specific tools for enhanced capabilities

🔹 Comprehensive Memory Model

Semantic Memory: Vector-based retrieval using Qdrant
Episodic Memory: Learning from past experiences
Procedural Memory: Encoding business SOPs into graph logic

🚀 Quick Start

🔹 Prerequisites

Ensure you have Docker and Docker Compose installed.

🔹 Step 1: Environment Configuration

Create a .env file in the root directory and add:

OPENROUTER_API_KEY=your_api_key_here

🔹 Step 2: Build and Run

cd infra
docker-compose up --build -d

🔹 Step 3: Access & Testing

Interactive API Docs: http://localhost:8000/docs Stopping the system:

docker-compose down

🗄️ 5. Database Management (pgAdmin)

🔹 Connecting to the Database

Field	Value
Host	db-persistence (or localhost if using host)
Port	5432
Maintenance Database	ai_memory
Username	user
Password	password

🔮 6. Extended Future Roadmap

🔹 Modular Toolsets

Financial analysis tools
Market intelligence tools

🔹 Cognitive Memory Tiers

Semantic Memory: Vectorized knowledge via Qdrant
Episodic Memory: Learning from historical interactions
Procedural Memory: Business SOPs embedded into graph logic

Project Structure

C:.
│   .gitignore
│   README.md
│
├───backend
│   │   .env
│   │   requirements.txt
│   │
│   ├───app
│   │   │   main.py
│   │   │
│   │   ├───api
│   │   │   │   deps.py
│   │   │   │
│   │   │   └───routes
│   │   │       │   chat.py
│   │   │       │   health.py
│   │   │       │
│   │   │       └───__pycache__
│   │   │               chat.cpython-311.pyc
│   │   │
│   │   ├───core
│   │   │       config.py
│   │   │       logger.py
│   │   │
│   │   ├───schemas
│   │   │   │   agent.py
│   │   │   │   chat.py
│   │   │   │
│   │   │   └───__pycache__
│   │   │           chat.cpython-311.pyc
│   │   │
│   │   └───__pycache__
│   │           main.cpython-311.pyc
│   │
│   └───engine
│       │   README.md
│       │   state.py
│       │
│       ├───agents
│       │   │   base.py
│       │   │
│       │   ├───ceo
│       │   │   │   agent.py
│       │   │   │   config.yaml
│       │   │   │   prompt.txt
│       │   │   │
│       │   │   └───__pycache__
│       │   │           agent.cpython-311.pyc
│       │   │
│       │   ├───chro
│       │   │   │   agent.py
│       │   │   │   config.yaml
│       │   │   │   prompt.txt
│       │   │   │
│       │   │   └───__pycache__
│       │   │           agent.cpython-311.pyc
│       │   │
│       │   ├───manager
│       │   │   │   agent.py
│       │   │   │   config.yaml
│       │   │   │   prompt.txt
│       │   │   │
│       │   │   └───__pycache__
│       │   │           agent.cpython-311.pyc
│       │   │
│       │   ├───orchestrator
│       │   │   │   agent.py
│       │   │   │   config.yaml
│       │   │   │   prompt.txt
│       │   │   │   routerDecision.py
│       │   │   │
│       │   │   └───__pycache__
│       │   │           agent.cpython-311.pyc
│       │   │           orchestrator.cpython-311.pyc
│       │   │           routerDecision.cpython-311.pyc
│       │   │
│       │   ├───SummarizerAgent
│       │   │   │   agent.py
│       │   │   │   config.yaml
│       │   │   │   prompt.txt
│       │   │   │
│       │   │   └───__pycache__
│       │   │           agent.cpython-311.pyc
│       │   │
│       │   └───__pycache__
│       │           base.cpython-311.pyc
│       │
│       ├───configs
│       ├───llm
│       │   │   client.py
│       │   │
│       │   ├───providers
│       │   │   │   groq.py
│       │   │   │   openRouter.py
│       │   │   │
│       │   │   └───__pycache__
│       │   │           groq.cpython-311.pyc
│       │   │           openRouter.cpython-311.pyc
│       │   │
│       │   └───__pycache__
│       │           client.cpython-311.pyc
│       │
│       ├───memory
│       │   │   long_term.py
│       │   │   short_term.py
│       │   │
│       │   └───__pycache__
│       │           short_term.cpython-311.pyc
│       │
│       ├───orchestrator
│       │   │   orchestrator.py
│       │   │   router.py
│       │   │
│       │   └───__pycache__
│       │           orchestrator.cpython-311.pyc
│       │
│       ├───scripts
│       ├───supervisor
│       ├───tests
│       ├───tools
│       │   │   registry.py
│       │   │
│       │   └───__pycache__
│       │           registry.cpython-311.pyc
│       │
│       └───__pycache__
│               state.cpython-311.pyc
│
├───frontend
└───infra
        docker-compose.yml
        Dockerfile.backend
        Dockerfile.frontend