エピソード

  • How Mobile Apps Use On-Device AI for Real-Time Sign Language Translation

    How Mobile Apps Use On-Device AI for Real-Time Sign Language Translation
    2026/07/21
    Lucas and Luna explore a compelling new use case for on-device AI: real-time sign language translation via mobile apps. They dive into how companies like SignAll and Google are training models on fingerspelling datasets, the challenge of capturing facial expressions and body movement alongside hand gestures, and why Apple's Neural Engine makes low-latency inference possible without cloud dependency. They also discuss the accuracy trade-offs, the importance of preserving grammatical structure in sign languages like ASL, and what this technology means for accessibility. Specific numbers: the model runs at under 100 milliseconds per frame, uses about 2 watts of power, and supports a vocabulary of roughly 500 signs in current consumer apps. The conversation closes with the question of whether on-device translation can ever capture the nuance of a native signer. #SignLanguageTranslation #OnDeviceAI #MobileTechnology #Accessibility #ASL #NeuralEngine #SignAll #Google #MachineLearning #iOS #Android #RealTimeTranslation #DeepLearning #GestureRecognition #EdgeAI #FexingoBusiness #BusinessPodcast #Technology Keep every episode free: buymeacoffee.com/fexingo
    続きを読む 一部表示
    12 分
  • How Mobile Apps Recognize Handwritten Notes With On-Device AI

    How Mobile Apps Recognize Handwritten Notes With On-Device AI
    2026/07/21
    Lucas and Luna explore how mobile apps now use on-device AI to recognize handwriting in real time, no cloud needed. Starting with Google's Handwriting Input API and the MyScript engine (now part of Apple's PencilKit), they dive into the technical challenge: training models on millions of handwritten samples across dozens of scripts. They discuss how Apple's 2025 on-device Transformer architecture improved cursive recognition by 40 percent, and how apps like GoodNotes and Notability use custom models to convert scribbles into searchable text. The conversation covers the privacy benefits — your handwriting data never leaves the phone — and the surprising accuracy rates: up to 97 percent for clean print, 92 percent for messy cursive. Luna shares a demo fail anecdote from a developer conference where the model confused her recipe for 'spaghetti carbonara' as 'spaghetti carbonara' (it was correct, but the audience laughed). They wrap up with a forward look: handwriting-to-text is now being integrated into more note-taking and productivity apps, potentially killing standalone OCR scanners. #HandwritingRecognition #OnDeviceAI #MobileApps #ApplePencilKit #MyScript #GoogleHandwritingInput #GoodNotes #Notability #Privacy #Transformer #CursiveRecognition #OCR #NeuralEngine #MachineLearning #TechPodcast #Technology #FexingoBusiness #BusinessPodcast Keep every episode free: buymeacoffee.com/fexingo
    続きを読む 一部表示
    9 分
  • How Mobile Apps Are Digitizing Your Physical Receipts With On-Device AI

    How Mobile Apps Are Digitizing Your Physical Receipts With On-Device AI
    2026/07/20
    Lucas and Luna dive into the world of receipt-scanning apps that use on-device AI to extract, categorize, and analyze purchase data without sending your shopping history to the cloud. They break down how models like Apple's Vision framework and open-source OCR engines parse messy till slips, handle faded thermal paper, and classify spending categories in real time. They also discuss the privacy angle: why keeping receipt data on your phone matters for financial tracking apps, and how one startup, 'ReceiptHero', processes over 50 million receipts per month entirely on-device. Plus, they explore the technical challenges — skewed images, multilingual text, and handwritten notes — and what this means for the future of personal finance automation. A focused, numbers-driven conversation for anyone curious about the invisible AI turning paper clutter into digital insight. #OnDeviceAI #ReceiptScanning #MobileApps #ComputerVision #OCR #AppleVision #Privacy #PersonalFinance #ExpenseTracking #MachineLearning #RealTimeProcessing #ThermalPaper #ReceiptHero #EdgeAI #CoreML #Technology #FexingoBusiness #BusinessPodcast Keep every episode free: buymeacoffee.com/fexingo
    続きを読む 一部表示
    10 分
  • Mobile Apps That Identify Edible Wild Mushrooms With On-Device AI

    Mobile Apps That Identify Edible Wild Mushrooms With On-Device AI
    2026/07/20
    This episode explores how mobile apps use on-device AI to identify edible wild mushrooms in real time. Lucas and Luna dive into the technical challenges of training models on thousands of mushroom species, the safety implications, and how apps like Picture Mushroom and Mushroom Identifer achieve high accuracy without cloud reliance. They discuss the dataset size, model compression, and the importance of caution when foraging. #OnDeviceAI #MobileApps #MushroomIdentification #WildEdibleMushrooms #Foraging #ComputerVision #ImageRecognition #MobileDevelopment #iOS #Android #MachineLearning #AI #Technology #FexingoBusiness #BusinessPodcast #TechPodcast #LucasAndLuna #RealTimeAI Keep every episode free: buymeacoffee.com/fexingo
    続きを読む 一部表示
    9 分
  • How Mobile Apps Use On-Device AI for Real-Time Music Source Separation

    How Mobile Apps Use On-Device AI for Real-Time Music Source Separation
    2026/07/19
    Ever wondered how apps let you isolate vocals from a song or remove background noise from a recording — all on your phone, no cloud processing? In this episode, Lucas and Luna dive into music source separation, the AI technique behind the phenomenon. Lucas explains how modern on-device neural networks, trained on spectrograms, can split a mixed audio track into stems like vocals, drums, bass, and other instruments in real time. They dissect the architecture of small-footprint U-Net models that run efficiently on a phone's neural engine, compare performance across iOS and Android, and discuss the trade-offs between latency, quality, and battery drain. Luna brings up real-world use cases beyond karaoke — from podcast production to hearing aid filtering — and challenges Lucas on whether the hype around on-device separation matches the current reality. They also touch on licensing gray areas. By the end, you'll understand exactly how your phone can 'unmix' a song in seconds. #MusicSourceSeparation #OnDeviceAI #MachineLearning #NeuralNetworks #AudioProcessing #DeepLearning #iOS #Android #MobileApps #Technology #FexingoBusiness #BusinessPodcast #RealTimeAI #Spectrograms #UNet #StemSeparation #KaraokeApps #PodcastProduction Keep every episode free: buymeacoffee.com/fexingo
    続きを読む 一部表示
    7 分
  • How Apps Detect Mushroom Species With On-Device AI

    How Apps Detect Mushroom Species With On-Device AI
    2026/07/19
    Imagine pointing your phone at a wild mushroom and getting an instant species ID, toxicity warning, and lookalike alert — all processed on the device with no internet connection. In this episode of Mobile Development with Fexingo, Lucas and Luna explore how modern on-device AI makes real-time mushroom identification possible. They break down the core technical stack: a fine-tuned MobileNetV3 running on Core ML and TensorFlow Lite, trained on a dataset of over 3,000 European and North American species with 500,000+ labeled images. They discuss the challenge of distinguishing deadly Amanita phalloides from edible Volvariella volvacea, the trade-offs between accuracy and model size, and why Apple's Neural Engine and Android's NNAPI make this feasible at 30 frames per second. Lucas also explains the safety-first approach: the app never says 'safe to eat' — only provides a probability score and a list of lookalikes. The episode lands on the broader trend: niche, high-stakes AI apps that would have required a cloud server five years ago now run entirely offline on a phone. #MushroomIdentification #OnDeviceAI #MobileApps #CoreML #TensorFlowLite #ComputerVision #MachineLearning #Fungi #iOS #Android #AppleNeuralEngine #NNAPI #DeepLearning #MobileNetV3 #ToxicityDetection #CitizenScience #FexingoBusiness #BusinessPodcast Keep every episode free: buymeacoffee.com/fexingo
    続きを読む 一部表示
    11 分
  • How Mobile Apps Use On-Device AI for Real-Time Car Damage Assessment

    How Mobile Apps Use On-Device AI for Real-Time Car Damage Assessment
    2026/07/18
    In this episode of Mobile Development with Fexingo, Lucas and Luna explore how on-device AI is transforming car damage assessment for insurance claims and repairs. They dive into the technology behind apps like Tractable's AI, which can analyze photos of a dent or scratch and estimate repair costs in seconds—all on the phone without sending data to the cloud. Lucas explains the machine learning models that detect and classify damage types (scratches, dents, cracks) and how they achieve low latency while preserving user privacy. Luna discusses the practical challenges, such as handling varied lighting conditions and angle distortions, and how developers are training models with synthetic damage images to overcome data scarcity. They also touch on the impact for insurers, who can now process claims faster, and for body shops, who get instant quotes. The episode wraps with a forward-looking question about whether this technology could eventually bypass human appraisers entirely. A light donation mention ties the value of the conversation to listener support. #OnDeviceAI #MobileAppDevelopment #CarDamageAssessment #InsuranceTech #Tractable #AI #MachineLearning #ComputerVision #RealTimeAI #ClaimsProcessing #DeepLearning #SyntheticData #Privacy #Latency #AutoRepair #Technology #FexingoBusiness #BusinessPodcast Keep every episode free: buymeacoffee.com/fexingo
    続きを読む 一部表示
    7 分
  • How Mobile Apps Use On-Device AI for Real-Time Plant Disease Detection

    How Mobile Apps Use On-Device AI for Real-Time Plant Disease Detection
    2026/07/18
    Lucas and Luna explore how modern smartphones can diagnose plant diseases in real time using on-device AI. They dive into a specific case: a 2025 study from the Indian Institute of Technology where a lightweight convolutional neural network achieved 97.3 percent accuracy identifying late blight in tomato leaves, running entirely on a mid-range phone without internet. The hosts discuss the model architecture, the training data challenge (over 50,000 labeled images), and why deploying AI on-device matters for farmers in rural areas with limited connectivity. They also touch on Apple's Core ML and Google's TensorFlow Lite as the enabling frameworks, and how this technology is starting to appear in apps like PlantVillage and Plantix. The conversation closes on whether consumer apps can maintain accuracy across thousands of different crop diseases. #OnDeviceAI #PlantDiseaseDetection #MobileAI #DeepLearning #AgricultureTech #CNN #TensorFlowLite #CoreML #IIT #TomatoLateBlight #PlantVillage #Plantix #MobileComputing #EdgeAI #SmartFarming #Technology #FexingoBusiness #BusinessPodcast Keep every episode free: buymeacoffee.com/fexingo
    続きを読む 一部表示
    11 分