✈️ ActuAI: Aerospace Production Workflow Automation

📚 Full project documentation — architecture, the five missions in depth, security model, deployment and verification: docs/PROJECT.md

📖 Project Overview

ActuAI is a secure, multi-agent artificial intelligence system designed to automate Non-Value Added (NVA) administrative tasks within a highly regulated aerospace Actuation service. Operating under strict EN9100 compliance constraints, the application bridges the gap between unstructured communication (e.g., supplier emails) and rigid enterprise resource planning (ERP) systems (such as SAP).

To guarantee data sovereignty and confidentiality, the core of the system is designed to run On-Premise (Edge Infrastructure) with "Human-in-the-Loop" (HITL) validation checkpoints.

🎯 Automated Service Missions

The Actuation service acts as the operational backbone for thrust reverser manufacturing. ActuAI aims to streamline the following five core missions by eliminating manual data entry, double-checking, and scattered information retrieval:

1. Component Supply Chain Monitoring: Automating the extraction of delivery statuses, delays, and tracking information from daily supplier emails to continuously update the ERP system without double data entry.
2. Production Schedule Coordination: Proactively monitoring discrepancies between predicted ERP delivery dates and actual supplier updates to flag potential assembly line blockages before they impact production.
3. Quality and Non-Conformance Management: Automatically pre-filling Non-Conformance Reports (NCR/FNC) and tracking 8D corrective action reports by retrieving existing metadata directly from the ERP, significantly reducing manual drafting time.
4. Technical Documentation Control: Streamlining the compilation and version control of mandatory manufacturing records by aggregating data spread across network drives, ERPs, and email archives.
5. End-to-End Component Traceability: Creating a unified, instantly searchable context that reconstitutes the complete history of a component from initial order to final integration, fulfilling strict aerospace traceability requirements.

🛠️ Core Tech Stack

• Agent Orchestration: LangGraph & LangChain
• Backend & API: FastAPI, SQLModel, Pydantic
• Frontend UI: React (Human-in-the-loop Dashboard)
• Dependency Management: uv (Monorepo Workspace)
• Data Storage: PostgreSQL (Structured Data) & Qdrant (Vector Database for RAG)
• Deployment: Docker & Docker Compose

---

📂 Repository Structure (Monorepo)

The project is structured as a Monorepo managed by uv workspaces, containing three decoupled microservices:

ActuAI/
├── pyproject.toml              # Global workspace configuration
├── uv.lock                     # Global dependency lockfile
├── .env                        # Global environment variables
├── docker-compose.yml          # Infrastructure orchestration (Databases)
│
├── actuai_mock_data/           # 🏭 MODULE 1: Industrial Simulation
│   # Simulates the air-gapped SAP ERP (BAPI) and generates unstructured 
│   # supplier emails, Excel dashboards, and PDF technical documents.
│
├── actuai_backend/             # 🧠 MODULE 2: Core Logic & AI Orchestration
│   # Contains the ETL pipelines, the PostgreSQL/Qdrant connectors, 
│   # and the LangGraph Multi-Agent system (Supervisor, Transactional, Investigative).
│
└── actuai_frontend/            # 💻 MODULE 3: User Interface
    # A comprehensive React-based frontend providing the Human-in-the-Loop 
    # validation dashboard for aerospace experts to review AI-drafted actions.

---

⚙️ Installation & Local Development

We use uv for ultra-fast dependency management and virtual environment resolution across the entire workspace.

1. Prerequisites

• Python >= 3.13
• Docker & Docker Compose
• uv package manager (curl -LsSf https://astral.sh/uv/install.sh | sh)

2. Initial Setup

Clone the repository and set up your environment variables:

git clone <repository_url>
cd ActuAI
cp .env.example .env  # Ensure you configure your local paths and API keys

Synchronize the entire workspace. This single command will create the .venv at the root and install dependencies for all sub-projects:

uv sync

3. Launching the Local Infrastructure (Datalake)

Before running the applications, start the backend databases (PostgreSQL and Qdrant vector store):

docker compose up -d

• *Qdrant Dashboard available at: http://localhost:6333/dashboard*

4. Running the Microservices for Development

A. Start the SAP Mock API:

uv run uvicorn actuai_mock_data.sap_api.main:app --port 8080 --reload

• Generate initial fake documents/emails: uv run python actuai_mock_data/generators/main.py

B. Start the Backend (LangGraph & ETL):

# In a new terminal tab — run from the backend's import root (src/)
cd actuai_backend/src
uv run uvicorn main:app --port 8000 --reload

C. Start the Frontend (React HITL dashboard):

# In a new terminal tab
cd actuai_frontend
npm install
npm run dev   # Vite dev server; proxies /api to the backend on :8000

---

🐳 Docker Deployment

The root docker-compose.yml brings up the data layer (PostgreSQL + Qdrant), the simulated SAP ERP (actuai_mock_data) and the orchestrator backend in one command:

docker compose up --build -d

This starts:

• actuai-postgres (:5432) and actuai-qdrant (:6333/:6334) — the datalake
• actuai-mock-data (:8080) — the simulated SAP BAPI (seeded with Faker data)
• actuai-backend (:8000) — the FastAPI orchestrator (USE_MOCK_LLM=true, ETL_AUTO_START=true, so it pulls from the mock SAP on boot)

The React frontend is run on the host with Vite (cd actuai_frontend && npm run dev), whose proxy forwards /api to the backend published on :8000.