Freelance Full-Stack Developer|CTO @ CaLab.AI |ML Developer & Data Analyst| Compter Science Student @ University of Calabria | Change Model Specialist @ UNIDEA |Community Lead @ Velletri-Dev | Author
Daniele Mario Areddu is a Computer Science student at the University of Calabria and a freelance developer. He works on applied Artificial Intelligence projects, focusing on the fine-tuning of language models and the development of Retrieval-Augmented Generation (RAG) systems to enable more contextual and intelligent human–machine interactions. In the field of web development, he builds full-stack applications using Angular for the front-end and Node.js or Python(API REST,Flask etc.etc.) for the back-end, with a strong emphasis on scalability, maintainability, and user experience. He is also passionate about game development, creating Minecraft plugins and mods in Java, and prototyping games with C++ in Unreal Engine, where he integrates AI-driven NPC behavior and advanced gameplay mechanics. A tech enthusiast and speaker, Daniele frequently participates in workshops, conferences, and events focused on digital innovation and AI. He is the community lead of Velletri-Dev, organizing meetups and events for students and developers in the Velletri area. He also collaborates with UNIDEA as a Change Model Specialist, promoting digital transformation and a culture of innovation Since May 2025 he has been collaborating with CaLab.Ia, an innovative company in the world of AI, as Director of the technical department.
Talk · 1
Questo talk presenta un’architettura edge-first per Flutter che sfrutta Gemini Nano (on-device) e Gemini (cloud) per costruire applicazioni AI-native con fallback dinamico e latenza controllata. L’obiettivo è dimostrare come implementare una pipeline vocale completa interamente in Dart, con FFI e gestione reattiva dello stato. A partire da un’app che sto sviluppando – NoteVoice AI – mostreremo come: acquisire input audio con flutter_sound, eseguire trascrizione offline tramite Whisper.cpp usando Dart FFI o flutter_rust_bridge, serializzare il testo trascritto in prompt ottimizzati per Gemini Nano, integrato tramite Android ML System API o plugin custom su device compatibili, eseguire inference LLM locale su dispositivi Pixel o fallback automatico verso Gemini via API REST in caso di limitazioni hardware o prompt complessi. L’architettura dell’app segue un pattern AI Manager con: controllo automatico della capacità di calcolo, routing delle richieste su base contestuale (locale vs cloud), astrazione unificata per le risposte, in modo da mantenere l’UI consistente indipendentemente dalla sorgente AI. Il talk include benchmark (latency, RAM, CPU) e un confronto tra: Flutter vs React Native per binding nativi e gestione modelli AI, modelli quantizzati GGUF vs Gemini Nano, e integrazione FFI in contesti real-time. Chiuderemo con riflessioni su privacy, resilienza offline e design AI-centrico. Stack: Flutter 3.22+, Dart FFI, Riverpod Whisper.cpp (C++) --> Dart Gemini Nano (Android System API) Gemini Pro (REST API JSON) Conditional fallback + prompt context chaining