Diletta Contarin
Automotive Diagnostics from Natural Speech: A Retrieval-Augmented Generation Approach.
Rel. Luca Cagliero. Politecnico di Torino, Corso di laurea magistrale in Data Science And Engineering, 2026
|
|
PDF (Tesi_di_laurea)
- Tesi
Accesso limitato a: Solo utenti staff fino al 24 Luglio 2027 (data di embargo). Licenza: Creative Commons Attribution Non-commercial No Derivatives. Download (2MB) |
Abstract
Vehicle diagnostics currently rely on technicians manually transcribing spoken symptom descriptions and cross-referencing them with manufacturer-specific service manuals, a procedure representing a considerable portion of their working time per diagnostic report. This workflow is error-prone and inefficient, and is made worse by the vocabulary mismatch between the colloquial language used by vehicle owners and the formal terminology of technical documentation. Existing diagnostic systems focus on direct trouble code reading rather than natural language symptom interpretation, and do not address this semantic gap, leaving a significant unmet need in automated support for preliminary diagnostics. This thesis presents an end-to-end pipeline integrating Automatic Speech Recognition (ASR), Natural Language Processing (NLP), and Retrieval-Augmented Generation (RAG) to automate the conversion of spoken vehicle problem descriptions into grounded diagnostic responses with page-level citations, optionally delivered as synthesised speech.
The system employs Whisper Small for speech transcription and a locally deployed large language model for privacy-preserving response generation
Relatori
Anno Accademico
Tipo di pubblicazione
Numero di pagine
Corso di laurea
Classe di laurea
Aziende collaboratrici
URI
![]() |
Modifica (riservato agli operatori) |
