TL;DR: My journey with bE-More started in february 2025, when my team started to gear up for the Project Day, the most anticipated event in my school. Every year this contest transforms the school in a tech hub, where teams from all classes gather to showcase their projects to representatives of com
TL;DR: My journey with bE-More started in february 2025, when my team started to gear up for the Project Day, the most anticipated event in my school. Every year this contest transforms the school in a tech hub, where teams from all classes gather to showcase their projects to representatives of companies of our territory. To be sure we developed a project that would be in demand, we knew we had to resolve a real-world problem. That's how we decided to look at Agenda 2030, which challenges companies to comply with stricter environmental standards and enhance their energy savings. This is why we decided to rely on a local AI analysis system. By using Ollama and the (then) newborn Mistral:7b, we created a light-weight, fast and secure solution. It allows users to get data insights and detect anomalies without spending hours reading long, complex charts. Our solution relies on different components talking to each other to gather, store and analyze data for the end user, who can then make decisions based on AI-driven insights. The IoT Hub (ThingsBoard): We selected platform for robust data handling and storing, using MQTT and real-time dashboards on the port 8080. The Edge Hardware (Arduino): The actual operator of the system. It is responsible for gathering real-time telemetry from light sensors and executing local automation rules to save energy. The Intelligence (Python & Ollama): A middleware service that acts as a bridge between the data and the AI model by querying the PostgreSQL database of ThingsBoard to feed the Mistral:7b model. To be able to communicate, the components used MQTT and HTTP(s) protocols, which make the architecture fast and secure. MQTT (Message Queuing Telemetry Transport): a light-weight transport protocol to exchange messages based on the subdivision of devices in publishers and subscribers, relying on a central broker that rules the exchange of telemetries in channels called topics. In our project, it's the heartbeat of the communication between the Arduino and ThingsBoard. HTTP (HyperText Transport Protocol): a request/response protocol, relying on different types of calls (i.e. GET, POST, PUT). We used it to serve the web-based dashboards from the server to the Java app. While HTTP is natively insecure, it is possible to implement HTTPS to encrypt data between the server and the end-user, ensuring that corporate sensitive data remains protected from interceptions. To demonstrate bE-More during the showcase, we didn't just show a screen with code, we built a physical prototype. This diorama represents a typical office environment where our architecture could be tested in real-time. The diorama was designed to be a miniature office. We integrated the sensor and actuators directly into the structure to simulate a real-life installation: The Main Workspace LEDs: Representing the lighting system, controlled by the Arduino. Status Feedback: A dedicated LED is used to indicate if the automatic mode is activated, while a Buzzer is used to signal acoustically when the mode is switched. Interactive Controls: Two physical buttons were placed on the Arduino Shell to allow the user to toggle between manual control and Autonomous Mode. IV.II Circuitery in the Crafted Arduino Shell The central point of the diorama is an Arduino microcontroller (Arduino UNO R4, to natively host Wi-Fi functionalities) that runs a deterministic logic to ensure energy saving without human intervention. The automation follows a simple rule: Sensor Input: a photoresistor (connected to pin A3) continuously monitors the environments natural light. The Publishing: the Arduino automatically logins (registering can be both manual and automatic on the platoform) to the ThingsBoard MQTT broker, which stores data in the database of the platform. The Threshold: if the 'AUTO' mode is active and the natural light value exceeds a value of 450, the system automatically powers down the lighting system. V. Conclusion: The Victory Thanks to the project and development process, where we found great synergies between the selected technologies, we developed a working product that could be installed in a real-life scenario. The Project Day 2025 was an unforgettable experience, where the interest of the company representatives was the ultimate validation of our hard work. We were honored to be announced as winners, a title we shared with another brilliant project from our class (VIPRA, whose github repo can be found here). This double victory proved that our school is a real nursery for innovation, where sustainability and technology meet to solve actual problems. 1. Security (HTTPS) While the initial web server used HTTP, a professional migration to HTTPS would be the next priority. By adding an SSL/TLS layer, we can encrypt all data between the server and the end-user, ensuring total privacy during delivery. 2. Artificial Intelligence The local AI landscape has exploded since we first started. We could now enhance our analysis through: ↳ Prompt Engineering: refining the input prompts to make more accurate insights from telemetry. ↳ Llama 3: swapping to a more modern model to use better reasoning capabilities while keeping the same infrastructure. ↳ Ad Hoc AI Models: creating a custom-trained model based on specific analysis samples, allowing a fine-tuned weight balancing. 3. Actuators Decoupled Architecture: in the diorama, one Arduino handled both the sensor and actuator roles. In a real-life deployment, these would appear in separate devices distributed across different zones. ESP32 Board: for a final product, we would switch to ESP32 board, which include native Wi-Fi, smaller size and superior processing power. 4. Advanced MQTT Infrastructure Currently, our implementation uses an unilateral MQTT flow for telemetry. Introducing an intermediary broker, such as Mosquitto, we would enable bi-directional communication, allowing for real-time analysis and immediate automated correction through actuators without the need to risk database integrity and injections. GitHub (code) ThingsBoard Arduino MQTT HTTP Ollama Mistral:7b GiZano: I am a BSc in CS at University of Pisa with a deep passion for IoT, local AI, and sustainable tech. As the lead architect of bE-More, I focus on building systems that bridge the gap between complex software and real-world environmental impact - always with a privacy-first mindset. Beyond winning the Project Day 2025, I'm constantly exploring new ways to integrate modern stacks like Java 23, Python aiohttp and local LLMs into everyday life, Let's connect! Linkedin GitHub Personal website This article was originally published on my Medium blog
Key Takeaways #
- •TL;DR: My journey with bE-More started in february 2025, when my team started to gear up for the Project Day, the most anticipated event in my school. Every year this contest transforms the school in a tech hub, where teams from all classes gather to showcase their projects to representatives of com
- •This story was reported by Dev.to , covering developments in thedev space.
- •AI advancements continue to reshape industries — read the full article on Dev.to for complete coverage.
📖 Continue reading the full article: