HunarmandAI: A Bilingual Intelligent Diagnostic Interface for Smart Vehicles and Industrial Mechatronic Systems

Modern smart vehicles and industrial mechatronic systems generate complex sensor telemetry and alphanumeric fault codes that require accurate interpretation to ensure reliable and safe operation. In multilingual industrial environments, diagnostic documentation is predominantly available in English, while technicians rely on code-mixed vernacular languages, creating significant barriers to effective human-machine interaction and maintenance decision-making. Existing diagnostic systems primarily focus on general natural language processing or simple retrieval approaches and often fail to provide reliable interpretation of technical telemetry and multilingual maintenance queries. This paper introduces HunarmandAI, a bilingual AI-driven diagnostic interface designed for smart vehicle and industrial maintenance in Urdu-English contexts. We construct a domain-specific corpus of 48,975 samples (6.9 M tokens, 76.2% code-mixed) and propose a multi-stage Retrieval-Augmented Generation framework integrating multilingual dense retrieval, cross-encoder re-ranking, and Maximal Marginal Relevance diversity filtering. Evaluated on 200 expert-level diagnostic queries, the proposed Dense+re-ranking+MMR pipeline achieves an MRR of 0.742 and a faithfulness score of 0.730, significantly outperforming lexical, dense, and hybrid baselines. Results demonstrate improved diagnostic precision and grounded reasoning, establishing bilingual AI-based diagnostic interfaces as a practical and scalable solution for Industry 4.0 robotics and automation systems.