Use Cases and AI Projects

Some projects or details are omitted for NDA with customers or still WIP.
Please note: most projects involved other collaborators.

AI Agents

• Description: Developed AI agents and benchmarking tools integrating various LLM technologies
• Technologies:
  • Models & APIs: Hugging Face, Llama 3.x, Ollama, Llamafile (local execution), OpenAI API, Groq, Deep Seek
  • Frameworks: PhiData, CrewAI, MoA, LangChain, PydanticAI
• Implementation:
  • Developed custom software tools integrating LLM agents, including benchmarking of various AI models (to be released as open source)
    • Some experimental concepts applications available on Streamlit
  • Attended online courses from DeepLearning.ai and YouTube (AI Technologies details )

• Year: 2024-2025

RAG and Fine-Tuning

• Description: Implemented Retrieval-Augmented Generation (RAG) and fine-tuned LLM models with proprietary client datasets.
  
• Implementation:
  • Integrated private PDF datasets for domain-specific AI solutions.
  • Experimented with vector databases and embedding optimizations.
  • Fine-tuned LLMs to improve contextual accuracy (NDA-protected details).

• Year: 2023-2025

Benchmarking and Profiling Edge AI + TPUs

• Description: Benchmarking chip performances on AI tasks running on SoC, CPUs and TPUs accelerators
  
• Technologies HW Tools such as scopes/power meters; python and bash scripts
• Implementation:
  • Measured TOPS and TOPs/W (also using HW external scopes) on various HW platforms
  • defined and used common model for inference + vendors “optimization” sw tools
    • Typ AI models: Mobinelnet SSD v2 (Object Detection, Optimized), a CNN trained on COCO Dataset (80 obj) ; YOLO with various variants;
    • measures latency and accuracy on inferences

• Year: 2020-2025

Machine Learning Engineer

• Year: 2023-2025

Voice Recognition Offline Module for Artistic VR Project

• Description: Implemented an offline voice recognition engine for an artistic VR project in Unity.
• Technologies: Python, Unity (C#), Vosk (customized for slang/accent recognition).
  
• Implementation:
  • Configurable module maps activation words or short sentences to commands sent over LAN or a local server.
  • Collaborative project with Accademia Belle Arti Roma for “Non-Linear Story 2” VR metaverse experience.
  • Tested in both English and Italian.
  • Public link will be available upon release.
• Year: 2022

Prompt Engineering and ChatBot App (Interactive Theatre)

• Description: Developed a chatbot and prompt engineering tool optimized for LLM interaction, integrating real actors in interactive environments.
  
• Implementation:
  • Supports both local LAN and AWS deployment (accessible via public IP with firewall port forwarding).
  • Includes a collection of pre-built state-of-the-art LLM prompts with adjustable parameters.
  • Provides an option to fetch and update new prompts via GitHub.
  • Features custom hardware integration for interactive control of “Bit,” a 1990s-era AI persona from Museo delle Scienze Napoli, modernized with AI.
  • AI chatbot interface rendered in Unreal Engine (third-party development).
• Technologies: Python, FastAPI, Panel (GUI), OpenAI API (ChatGPT), open-source LLMs (Llama, Mistral), AWS Cloud.
  
• Hardware: Leap Motion devices and TouchDesigner (with Python AI callbacks) for UI control by human actors.
  
• Client: Bit-Bot.org (PM: Sintropia Consulting, Museum of Science - Napoli, Italy).
  
• Status: Experimental stage; public interactive workshops planned.
  
• Year: 2023

ChatBot “The Bilingual Politician Oracle”

“A political oracle that provides non-committal, open-ended answers.”
- Client: Matango Srl
- Technologies: Custom prompt engineering for OpenAI API (ChatGPT Plus).
- Live Demo: The Bilingual Politician Oracle (requires OpenAI API key).
- Year: 2023

Predictive Maintenance Project

• Description: Designed an AI-powered predictive maintenance solution to optimize maintenance schedules and reduce equipment failures.
  
• Implementation:
  • Developed ML models to analyze failure patterns in broadcasting hardware.
  • Leveraged AWS cloud for real-time data gathering and model inference.
• Technologies: Python, TensorFlow, Keras, GPT-4, AWS Cloud.
  
• Client: Elenos Srl (Radio & DTV hardware transmitters, deployed in 20K+ systems globally).
  
• Year: 2019

Robot Arm Tracker for Touch Screen Testing

• Description: Developed an automated robotic manipulator to test an industrial touchscreen interface for a Class B laser machine.
  
• Implementation:
  • Used OpenCV to track screen changes and detect crashes.
  • MCU-based robotic arm with servo motors to simulate human touch interactions.
  • 4-axis robotic motion planning with real-time feedback via HD webcam.
    
• Technologies: Python, C/C++, OpenCV, Arduino (Atmel MCU).
  
• Year: 2018

Object Detection System

• Description: Implemented an AI-powered object detection system for images and video streams.
  
• Implementation:
  • Trained models using cloud computing resources (Google Cloud, AWS).
    
  • Optimized deployment for local edge computing devices.
    
• Technologies: Python, TensorFlow, Keras, OpenCV, Roboflow (for data labeling).
  
• Year: 2016-2017

EEG Data Collector & Brain-Machine Interface (BMI)

• Description: Developed an EEG-based system for experimental Brain-Machine Interface (BMI) research.
  
• Implementation:
  • Collected EEG data using Muse BT devices.
  • Built custom signal processing models for brainwave pattern analysis.
  • Developed a cloud-based system for real-time EEG data processing and AI-powered interpretation.
    
• Technologies: Python, Muse BT API, AWS Cloud, FFT, ML signal processing.
  
• Client: SmartEcosystem Ltd (backed by Big Dream Ventures B.V).
  
• Year: 2017-2021

Electrical Signal Pattern Recognition

• Description: Developed AI-based pattern recognition models for electrical signal analysis and anomaly detection.
  
• Technologies: Python, API/Stacks, Llama 2, GPT-3x, GPT-4.
  
• Implementation: Focused on edge computing applications (except GPT-4, which required an internet connection).
  
• Year: 2017

Face Recognition Project

• Description: Built a face recognition system for identification and verification tasks.
  
• Implementation:
  • Trained the model on an Nvidia GTX250 GPU using a 20-person dataset.
  • Optimized for real-time inference on Raspberry Pi 3, 3+, and 4.
    
• Technologies: Python, OpenCV, SciKit-Learn, TensorFlow, Keras.
  
• Year: 2016

‘SAM’ Data Collector Project

• Description: Developed an Android & iOS app to collect and process sensor data for regression testing.
  
• Implementation:
  • Custom hardware with ADC connected via mic port for sensor data collection.
    
  • Cloud-based (Dropbox) metadata storage including GPS, device ID, and user ID (opt-in).
    
  • Python scripts for automated regression testing on Android (APK) and iOS (TestFlight).
    
• Year: 2016-2017