Lorenzo Siena's
portfolio

Cybershop is a fictional business management app set in the year 2079. Its purpose is to sell cybernetic prostheses to be displayed in the app through augmented reality or in 3D in the form of STL files, a CAD format, which once purchased can be printed at home using a 3D printer enabled for printing biocompatible implants and electronic circuits.

The project and demo were created for the subject Mobile Programming.

Technologies and Services Used:

• Firebase (Google Cloud Services) with:
  • Authentication: email management, user UID and authentication
  • Realtime Database: product list, users and wishlist
  • Storage: saving images and STL files of products

Images and texts were generated with Scribble Diffusion and ChatGPT. The demo is functional and responsive.

Repository and presentation available in PDF.

Go to the pdf Go to project

The project stems from the curiosity to explore the limits of the USB and MIDI protocol on a Novation Launchpad MK1.
Initially, through fuzzing with PyUSB and random packets, I managed to turn on the device's LEDs, even though it was not supported on Linux.

Subsequently, I delved into its internal workings: the Launchpad is a 9x9 grid of illuminable keys (3 levels of red and 3 of green, combinable into yellow), which responds to MIDI commands. I developed three Python scripts to describe the matrix, control its LEDs, and send systematic commands, using a wrapper that leverages amidi as a subprocess.

In addition to controlling the lights, the script receives and prints the MIDI events of the pressed keys, transforming the Launchpad into an interactive interface.

In this way I was able to turn the LEDs on, off, and color them as I pleased, without official drivers, even creating visual effects with random packets and instant shutdown sequences.

Go to video Go to project on GitHub Go to LinkedIn update

A tiny graphical interface for LAMPP, written in Bash and tput (mini graphical library). Its purpose is to start a local server without using the command line.

Go to project

Johnny The CyberCar Assistant is a proof of concept (PoC) of a fully local RAG-enabled AI agent grafted onto an Opel Corsa B from 1997, the result of my bachelor's thesis in computer engineering at the University of Catania.

Project Architecture

The project consists of 3 systems:

• A Linux server with a 1080 TI on which Cheshire Cat AI runs, the local model (LLAMA3), the voice synthesizer, and speech recognition (Whisper AI).
• A client mounted in the car (Esp32) with a 4g modem, a microphone, speaker and an Edge speech recognition module with a small machine learning model.
• The ESP32 interfaces with CANBUS to send and receive vehicle data

Description and Functionality

Inspired by KITT from Supercar (aka Knight Rider ) , Johnny makes the car smart: thanks to an ESP32 connected to the CAN BUS and with a 4g modem it enables a voice interface, long-term memory and intelligent control (via OBD-II Port) at the edge.
The rest of the AI system is executed remotely on a server with a consumer GPU (GTX 1080TI).

The assistant is invoked by the keyword "Hey Johnny!", after which it is possible to use voice commands to act locally on the car's devices (like raising the windows) or speak voice-to-voice remotely with the Chatbot, which responds with the context of the car's data (GPS position, speed, temperature, etc.), without ever taking your hands off the steering wheel.
The Qdrant vector database manages the assistant's short/long-term memory, enabling a true RAG (Retrieval-Augmented Generation) experience.

All services are orchestrated with Docker Compose.

Audio Pipeline

🎤 User voice
↓
❓ If the command is local → executes local action ⚙️
Otherwise:
↓
🐱 contacts johnny the remote chatbot
↓
🌐 Cloudflared (authentication)
↓
📝 WhisperAI (Voice2Text on GPU)
↓
🧠 LLaMA3 (Ollama, inference on GPU)
↓
🔊 Text2Speech (vocal response)
↓
❓ Loop: until it detects “stop”, returns to johnny 🐱

Examples of use

• "What's the weather like in Amsterdam?"
  ↓
  "Lorenzo, it's 23 degrees in Amsterdam."
• "Will I arrive in Berlin in time for dinner?"
  ↓
  "Yes Lorenzo, but you should refuel at the next station 33km from here."
• "I heard a strange noise, is something wrong with the car?"
  ↓
  "Lorenzo, from your CAN bus I read that you have a problem with the engine, I advise you to pull over as soon as possible."

N.B. The project is completely open-source, it was created in about 2 and a half months, it is designed for hardware reuse and edge computing in the automotive sector.
architecture, images, videos and full thesis in pdf (ITA ONLY) are available in the repository.

Components developed for the project

• Local Whisper Cat: An official Cheshire Cat AI plugin to transcribe locally on your gpu/cpu, in Docker.

  GitHub

• Chatty: Voice client in Python to contact the Cheshire Cat AI server.

  GitHub

• Chattino: Arduino/C++ voice client for ESP32 to contact the Cheshire Cat AI server.

  GitHub

Technologies used

Retromuseum is a university web development project for a fake e-commerce of a fictional company from Catania, where users could buy used games and consoles, with a visual style heavily inspired by classic 8-bit graphics.
Initially the backend was implemented in pure PHP, then migrated to Laravel.

It uses a pre-populated MySQL database, calls the Spotify API (OAuth 2.0) for the latest (fake) podcast available, and uses MongoDB as a database for the cart (only to get extra points for the project).

It includes a login system, with client/server validation, a search bar and dynamically generated content via JS from rawg.io api.

The entire web app is mobile-responsive.

Go to project