SAC vs. random
70.1%
Improvement over a random baseline in the two-building environment.
Open to Applied AI / ML roles - 2026
AI systems, built to be used.
Applied AI Engineer turning data, research, and language models into useful agents, prediction systems, and intelligent interfaces.
01Proof, not slogans
Measured results from real research work.
Outcomes from custom reinforcement-learning environments for building energy control, with agents benchmarked against a random baseline.
Peak-demand events
219 -> 0
Peak-demand events reduced from 219/720 to 0/720 control steps.
Custom environments
5
Progressively complex CityLearn environments designed for experimentation.
Applied projects
7+
Applied AI and ML projects spanning agents, RAG, analytics, and vision.
02What I build
Three areas where I turn models into working systems.
Focused systems that connect model capability to a real user, workflow, or research question.
AI Agents & RAG
Assistants, retrieval systems, document Q&A tools, and multi-tool workflows that reason over real data.
Data Analysis & Prediction
Dashboards, anomaly detection, forecasting workflows, machine learning pipelines, and intelligent analytics.
Applied AI & RL Research
CityLearn environments, SAC and PPO agents, energy optimization, and research-driven experimentation.
03Featured projects
Systems I've designed, built and shipped.
A selection spanning agents, retrieval, analytics, and computer vision. Source links are included where public.
/ 01Building nowAgent
RECALL-Agent
A self-improving multi-agent coding assistant that solves coding tasks, reflects on its failures and successes, stores structured lessons, and retrieves them on future tasks. Includes an evaluation harness comparing memory-on against memory-off, sandboxed code execution for honest reward signals, and measurable before/after learning results.
/ 02Analytics
AI Supply Chain Risk Dashboard
A Streamlit platform for predicting delivery risk, monitoring inventory, detecting anomalies, forecasting demand, and generating AI-assisted management reports.
/ 03Agent
AI Data Analyst Agent
A natural-language data assistant for cleaning datasets, answering questions, recommending charts, detecting anomalies, and training models.
/ 04Agent
Multi-Tool Research Agent
A research assistant that searches the web, arXiv, and Wikipedia, ranks source reliability, stores notes, and exports structured briefs.
/ 05Agent / Voice
AI Voice Receptionist
A free-friendly voice receptionist for small businesses. It greets callers, works out what they need one question at a time, collects lead details, detects urgency, decides whether to hand off to a human, summarizes the call, scores the lead as cold, warm, or hot, and saves everything to SQLite with a dashboard view. Runs fully free for demos, with an optional upgrade to real-time voice via LiveKit.
/ 06Agent / Voice
Bright Smile Dental Clinic Voice Agent
A real-time voice assistant for a dental clinic. It offers a free checkup, answers basic questions about the offer, collects appointment details, and saves booking requests. It does not give medical advice or diagnose. Testable through the browser mic or a real Twilio phone call over a bidirectional media stream.
/ 07Agent
CrewAI Business Agent System
A single-tenant AI operations workspace for small businesses and agencies. A protected Next.js dashboard on top of FastAPI, CrewAI, OpenAI embeddings, and persistent ChromaDB retrieval.
/ 08RAG
LlamaIndex Document Intelligence Platform
A SaaS-style document intelligence platform. Upload documents, index them, then chat over them with RAG where answers cite their sources. It also does structured field extraction, summarization, document comparison, and report generation.
/ 09RAG
PDF RAG Assistant
Document question answering with a local Chroma vector store, embedding fallback, and citation-aware answers with file and page references.
/ 10Agent / RL
RL Environment Auto-Designer
Describe a real-world optimization problem in plain English, such as optimizing hospital shift scheduling, and a pipeline of AI agents designs, codes, debugs, trains, and reports on a reinforcement learning environment for it.
/ 11Computer Vision
Facial Emotion Recognition
A CNN-based computer vision project trained on FER2013 to classify facial expressions including happy, sad, and angry.
/ 12Research
Research Navigator
A Streamlit application for discovering and previewing arXiv papers with NLP-powered search and summarization.
04Toolkit
The stack behind the systems.
Tools used across model development, agent workflows, research experiments, data products, and web interfaces.
Programming & Web07
- Python
- JavaScript
- HTML
- CSS
- React
- Next.js
- SQLite
AI Engineering07
- OpenAI API
- LangChain
- LangGraph
- LangSmith
- Hugging Face
- RAG
- Vector Databases
Machine Learning08
- PyTorch
- TensorFlow
- scikit-learn
- Stable Baselines3
- Computer Vision
- Reinforcement Learning
- SAC
- PPO
Data07
- Pandas
- NumPy
- Plotly
- Matplotlib
- Seaborn
- OpenCV
- Anomaly Detection
05Background
Experience & education.
Academic research and hands-on engineering across reinforcement learning, machine learning, and deep learning.
Experience
May 2026 - PresentNow
Research Intern, Reinforcement Learning
University of Verona, Italy
Mar 2026 - May 2026
Machine Learning Engineer Intern
University of Verona, Italy
Dec 2024 - Feb 2025
Deep Learning Intern
University of Verona, Italy
Education
2023 - PresentNow
MSc Artificial Intelligence
University of Verona, Italy
One-year course
Agentic AI
Krish Naik Academy, Edmonton
Current thesis research
Reinforcement Learning for Building Energy Management
Developing custom CityLearn environments with PV, battery, HVAC, and domestic hot water components, then evaluating SAC and PPO policies for energy optimization and peak-demand reduction.
06Contact
Let's build something useful.
Open to Applied AI Engineer, ML Engineer, AI Agent Engineer, and research collaboration opportunities. The fastest way to reach me is email.
Download CV