Thesis work - AI-Powered Fault Analysis System for Software-Defined Vehicle CCU

About this opportunity - Background

The central compute unit of modern software-defined vehicles generates complex fault reports that require extensive manual analysis. Current root cause analysis for these faults can range from hours to months, creating significant bottlenecks in the development and validation process. With approximately 200 fault reports per quarter requiring analysis, and each report containing megabytes to gigabytes of heterogeneous log data (DLT logs, Network traffic, coredumps, ...etc), engineers spend considerable time manually correlating information across multiple log formats to identify issues.
Additionally, the rapid evolution of the software platform means that behavioral patterns learned from historical data may become obsolete, requiring adaptive analysis approaches.

Scope of the thesis work

Primary Goal: 
Develop an AI-powered fault analysis system that reduces the time-to-diagnosis for fault reports from the vehicle's central compute unit.

Specific Objectives:
1.    Create specialized parsing and tokenization methods for automotive log formats (DLT, Pcap, coredump, blf)
2.    Build a Large Language Model-based system capable of analyzing multi-gigabyte fault report datasets
3.    Implement anomaly detection that identifies suspicious patterns when root cause analysis isn't possible
4.    Develop a confidence-aware system that distinguishes between identified root causes and observed anomalies
5.    Create a knowledge update mechanism to handle evolving platform behavior without full model retraining

Expected Deliverables:
•    A fine-tuned LLM specialized for automotive fault analysis
•    Custom tokenizers for DLT, Pcap, coredump and blf formats
•    Tool-augmented AI agent with access to debugging utilities (GDB integration for coredumps)
•    Knowledge update mechanism with retrieval support
•    Usable deployment with basic interface (stretch goal: cloud-deployed web service)
•    Evaluation framework with metrics on historical fault reports

What you'll bring

Essential Skills:
•    Strong Python programming (PyTorch/TensorFlow experience)
•    Understanding of transformer architectures and LLM fine-tuning
•    Linux/Unix systems programming knowledge
•    Version control (Git) and collaborative development
Preferred Skills:
•    Familiarity with automotive software development (QNX)
•    Experience with NVIDIA Drive platform or embedded GPU systems
•    Knowledge of debugging tools (GDB, core dump analysis)
Soft Skills:
•    Ability to work with domain experts to understand fault patterns
•    Strong analytical and problem-solving capabilities
•    Clear technical communication for cross-functional collaboration

Duration

•    The work will start in January 2026 
•    The duration for this thesis work is 20 weeks.
•    30 ECTS (Master academic credits) in agreement with your Thesis Advisor in University
•    This thesis is to be conducted by 2 Students working in pair.