ABSTRACT

Abstract−Coffee bean moisture content is a critical parameter that determines coffee quality and shelf life, with an idealstandard of 10-12.5%. CV Firm Kopi, a robusta coffee-processing MSME partner in Sidomulyo Village, Jember Regency, whose harvest has reached export markets, still relies on a manual method of biting the beans to estimate dryness during sun-drying, resulting in subjective and inconsistent measurements across workers. This condition makes it difficult for the partner to ensure consistent moisture-content quality when marketing its products. This community service activity aimed to implement a portable, machine learning-based coffee bean moisture meter to help the partner perform fast, non-destructive, and objective measurements. The device integrates a Capacitive Soil Moisture Sensor v1.2 and an ESP32 microcontroller with a Support Vector Regression model optimized using Gaussian Noise data augmentation (test R2 = 0.9760). The implementation method consisted of finalizing device calibration, designing a portable casing, internal testing, and on-site socialization and mentoring during the partner's coffee-drying process. The results showed that the partner's workers could easily operate the device after brief mentoring, and the measurement process became faster than the Grain Moisture Tester method because it required no hulling of the beans. The partner reported feeling more confident in determining the right time to stop the drying process. This activity is expected to be replicated at other coffee MSMEs to support qualitystandardization of Indonesian smallholder coffee.