Autonomous-Predictive-Maintenance

Vehicle Predictive Maintenance & Customer Interaction System

Overview

This project implements an end‑to‑end vehicle predictive maintenance and customer interaction system. It combines three intelligent agents working in a pipeline:

• Prediction Agent – Processes real‑time vehicle sensor data to detect anomalies.
• Diagnosis Agent – Analyzes anomalies using ML models + rule‑based logic to determine the likely issue.
• Customer Interaction Agent – Interacts with the vehicle owner in natural Hinglish, explains the issue, and schedules service appointments.

The architecture is modular, node‑based, and built using a LangGraph-style pipeline. Future agents like Scheduling, Insights, and Feedback can be plugged in easily.

---

Features

✅ Predictive Analysis

• Autoencoder-based latent representation of sensor data
• Reconstruction error to detect anomalies

✅ Explainable Diagnosis

• Rule-based domain logic
• XGBoost fallback model
• Optional FAISS-based RAG for contextual support
• Outputs structured JSON with:
  • diagnosis
  • confidence
  • explanation
  • source

✅ Human-like Customer Interaction

• LLM-powered Hinglish conversation
• Maintains conversational memory
• Supports MCP tool calls:
  • find_center
  • normalise_date
  • book_slot

  ✅ Human-like Customer Interaction

• LLM-powered Hinglish conversation
• Maintains conversational memory
• Supports MCP tool calls:
  • find_center
  • normalise_date
  • book_slot
• Dual interaction modes:
  • Text mode → traditional typed conversation
  • Voice mode → powered by Azure Cognitive Services Speech
    • Blocking TTS (agent speaks Hinglish in natural Indian voices)
    • Continuous STT (owner replies naturally, no timeout pressure)

✅ Simulation Ready

• Console-based interaction
• Can later integrate with Twilio, WhatsApp, or voice interfaces

✅ Security & Compliance (Future)

• UEBA for agent behavior monitoring
• Semantic Kernel for RBAC
• JSON-based structured inter-agent communication (MCP / A2A)

---

Folder Structure

├── CustomerInteraction/
│   ├── data/
│   └── tools/
|       |__tools.py
|       |__ tools_registry.py
│   ├── agent.py
│   ├── customer_node2.py
│   ├── node_base.py
│ 
│
├── Diagnosis_Agent/
│   ├── models/
│   │   ├── diagnosis_label_encoder.py
│   │   ├── diagnosis_node.py
│   │   ├── faiss.index
│   │   ├── faiss_meta.pkl
│   │   ├── node_base.py
│   │   ├── xgb_diagnosis.pkl
│   │   └── ...
│   └── tools/
│       ├── diagnosis_engine.py
│       ├── fallback.py
│       ├── generate_initial_csv.py
│       └── train_xg.py
│
├── Prediction_Agent/
│   ├── model/
│   │   └── models/
│   │       ├── autoencoder.pt
│   │       ├── scaler.pkl
│   │       ├── xgb_model.pkl
│   │       ├── classification_report.txt
│   │       ├── __init__.py
│   │       ├── node_base.py
│   │       ├── prediction_tool.py
│   │       ├── predictive_node.py
│   │       └── predictive_tool.py
│   └── train.py
│
├── insights_agent/
│   └── ... (future analytics agent)
│
├── schedulingAgent/
│   └── ... (future scheduling agent)
│
├── master.py
├── sensor_simulator.py
└── requirements.txt

--

Setup

1. Clone the repository

git clone <repo_url>

2. Environment Variables

Create a .env file: GROQ_API_KEY=<your_groq_api_key>

3. Run Pipeline

python Langraph_master.py

Future Agents

• Scheduling Agent – Auto-booking based on user preferences
• Insights Agent – Failure analytics + user behavior
• Feedback Agent – Post-service feedback collection

Tech Stack

• Python 3.10+
• NumPy, Pandas
• scikit-learn, XGBoost
• PyTorch (autoencoder)
• FAISS (RAG)
• Groq (LLM inference)
• MCPClient (tool calling)
• LangGraph (node orchestration)
• asyncio

Author

Advitiya Prakash