🚗 Car Diagnostic Assistant LLM

An AI-powered automotive diagnostic assistant built using Microsoft Phi-2, LoRA (Low-Rank Adaptation) fine-tuning, and Gradio. The system assists users by analyzing vehicle symptoms and OBD-II fault codes to generate structured diagnostic recommendations.

---

📖 Overview

Car owners often struggle to interpret vehicle symptoms and OBD-II trouble codes without consulting a mechanic. This project aims to bridge that gap by providing an intelligent diagnostic assistant capable of:

• Understanding vehicle symptoms described in natural language.
• Interpreting OBD-II fault codes.
• Suggesting likely causes.
• Assessing risk levels.
• Recommending safe checks and next actions.

The assistant combines a fine-tuned Large Language Model with curated automotive diagnostic knowledge to provide practical and easy-to-understand diagnostic guidance.

---

✨ Features

🔍 Symptom-Based Diagnosis

Analyze user-reported vehicle issues such as:

• Rough idling
• Engine vibration
• Hard starting
• Overheating
• Brake issues
• Electrical problems
• Transmission concerns

🛠 OBD-II Code Interpretation

Supports automotive fault code analysis including:

• P0300 – Random Misfire
• P0301 – Cylinder 1 Misfire
• P0302 – Cylinder 2 Misfire
• P0303 – Cylinder 3 Misfire
• P0304 – Cylinder 4 Misfire

and other diagnostic codes available through the integrated OBD database.

📊 Structured Diagnostic Reports

Generates outputs in a mechanic-friendly format:

• Symptom Summary
• OBD-II Interpretation
• Likely Causes
• Risk Level
• Safe Checks
• Do Not Do
• Next Action

🤖 Fine-Tuned LLM

Built using:

• Microsoft Phi-2
• PEFT LoRA fine-tuning
• Hugging Face Transformers
• PyTorch

💻 User-Friendly Interface

Interactive web application powered by Gradio.

---

🏗 System Architecture

User Symptoms
       +
   OBD Codes
       │
       ▼
 OBD Interpreter
       │
       ▼
 Prompt Builder
       │
       ▼
 Microsoft Phi-2
       +
  LoRA Adapter
       │
       ▼
 Output Validation
       │
       ▼
 Diagnostic Report

---

🧠 Model Information

Component	Details
Base Model	Microsoft Phi-2
Fine-Tuning Method	LoRA (PEFT)
Framework	Hugging Face Transformers
Training Dataset	Custom Automotive Diagnostic Dataset
Inference Engine	PyTorch
Interface	Gradio
Platform	macOS Apple Silicon / Local Deployment

---

📂 Project Structure

CAR_ASSISTANT/

├── assets/
│   ├── icons/
│   └── screenshots/
│
├── data/
│   ├── train.jsonl
│   └── obd-trouble-codes.csv
│
├── models/
│   └── car-assistant-qlora/
│
├── notebooks/
│   └── finetuning.ipynb
│
├── scripts/
│   ├── test_load.py
│   └── colab_parity_test.py
│
├── src/
│   ├── config.py
│   ├── model_loader.py
│   ├── obd_utils.py
│   ├── diagnosis.py
│   └── ui.py
│
├── app.py
├── requirements.txt
├── README.md
└── LICENSE

---

⚙️ Installation

Clone Repository

git clone https://github.com/YOUR_USERNAME/Car_Assistant_LLM.git

cd Car_Assistant_LLM

Create Virtual Environment

python -m venv venv

source venv/bin/activate

Windows:

venv\Scripts\activate

Install Dependencies

pip install -r requirements.txt

---

🚀 Running the Application

Launch the Gradio application:

python app.py

The application will be available locally at:

http://127.0.0.1:7860

---

🧪 Model Verification

To verify that:

• Phi-2 loads correctly
• LoRA adapter loads correctly
• Tokenizer is configured correctly
• Inference works successfully

run:

python scripts/test_load.py

---

📸 Screenshots

Home Interface

warnings Example

---

📈 Example Output

Symptom Summary:
- Strong fuel smell and engine vibration

OBD-II Interpretation:
- P0302: Cylinder 2 Misfire Detected

Likely Causes:
1. Faulty spark plug
2. Ignition coil issue
3. Fuel injector problem

Risk Level:
High

Safe Checks:
- Inspect spark plugs
- Check ignition connections

Do Not Do:
- Avoid prolonged driving

Next Action:
- Schedule engine diagnostic service

---

⚠️ Limitations

This project is intended for educational and research purposes.

• Not a substitute for professional vehicle diagnostics.
• Diagnostic recommendations are AI-generated and may not always be accurate.
• Supports a limited set of training scenarios and OBD fault patterns.
• Performance depends on the quality and completeness of user-provided information.

---

🔮 Future Improvements

• Expanded OBD-II code coverage
• Larger automotive training dataset
• Retrieval-Augmented Generation (RAG)
• Multi-turn conversational diagnostics
• Vehicle-specific recommendations
• Enhanced confidence scoring

---

👨‍💻 Project Team

Team Members

• Pranav M Nair
• Aadil Sandeep
• Advaith S Vinod
• Thejas Baiju

---

🎓 Academic Context

This project was developed as an academic and research-oriented exploration of:

• Automotive Artificial Intelligence
• Large Language Models
• Parameter-Efficient Fine-Tuning (PEFT)
• OBD-II Diagnostic Systems
• Intelligent Vehicle Assistance Systems

---

📜 License

This project is licensed under the MIT License.

See the LICENSE file for details.

---

🙏 Acknowledgements

• Microsoft Research for the Phi-2 model
• Hugging Face Transformers
• PEFT (Parameter-Efficient Fine-Tuning)
• PyTorch
• Gradio

---

Built with AI, Machine Learning, and Automotive Diagnostics in mind.