Ajlan AlAjlan

تخصص الذكاء الاصطناعي · دفعة 2026

Ovidio E-nose: TinyML for Scent Detection

Truffles, specialty coffee, and premium spices are foods of high value with a complex identity that is easy to imitate. This combination makes them frequent targets of fraud: a lower value item is sold as a premium one, or an aged sample is sold as fresh. Verifying such a claim today requires either expert judgment or a laboratory technique such as gas chromatography, and neither option is available at a point of sale in the field. We introduce Ovidio, a portable electronic nose that reads a food item’s scent and reports what it detects, without a laboratory. Ovidio combines an eleven channel gas sensor array and a BME680 environmental sensor, read by an Arduino UNO Q, into a complete scent classification pipeline that runs entirely on the device, with no dependence on a cloud connection. The system comprises OvidioBench, a public benchmark dataset of more than eighty two recording sessions; the Ovidio Classic Model (Legacy Model), a production classifier selected through a comparison of seven modeling methodologies and reaching a macro F1 score of 0.968 on a four class problem; a browser based workflow that lets a user retrain the classifier from newly recorded sessions without writing code; a dashboard that performs deterministic quality control; and a scent to speech stage that converts a classification into a spoken explanation grounded in the underlying sensor readings, produced by a language model and a speech synthesizer that both run on the device

اقرأ التقرير النهائي (PDF)