AI Engineer building production-grade agentic AI systems. Master’s student in Automation Engineering at the University of Bologna, specializing in Multimodal RAG, LLM orchestration, and intelligent automation.

• Portfolio • LinkedIn • Email • Resume

I’m a Master’s student in Automation Engineering at the University of Bologna (Italy) specializing in Multimodal RAG, LLM orchestration, and intelligent automation.

I treat AI models the same way I treat any other software component — they need clean interfaces, proper error handling, and a deployment story. My work focuses on turning messy data + complex workflows into reliable, observable, deployable systems.

I enjoy bridging research-grade AI concepts with real-world engineering, building systems that are:

• Modular
• Scalable
• Observable
• Actually deployable

• Multi-Agent Orchestration & Tool-Use Patterns
  Orchestrating multi-step agent ↔ tool ↔ UI flows with routing, retries/fallbacks, and observability hooks (traces, latency, failure tracking).

• Multimodal RAG & Retrieval Pipelines
  Retrieval pipelines across text/image/numeric signals with hybrid search + validation loops for real-world assistant use-cases.

• Large-Scale Data Clustering & ETL
  Clustering and grouping pipelines (K-means, DBSCAN, hierarchical) + preprocessing for product similarity, segmentation, and analytics.

• LLM Evaluation, Guardrails & Observability
  Practical eval and monitoring patterns for production assistants: rubric scoring, failure analysis, and system-level reliability.

• End-to-End ML Deployment on Cloud
  Shipping full-stack AI experiences from notebook to user, with reproducible workflows and deployable services.

• MemorAIz Onboarding Assistant – Split-screen conversational UI that auto-fills profiles using parallel LLM racing, hybrid caching, and streaming
• Fruugle Data Clustering – 1M+ product/pricing records preprocessing + clustering (K-means/DBSCAN/hierarchical) for comparable product segments
• Adversarial Training vs Domain Randomization – RL robustness in a modified LunarLander environment with adversarial self-play + domain randomization
• Adversarial Attacks & Defenses on CelebA – FGSM/PGD attacks on ResNet-18 with adversarial training and robustness analysis
• Multi-Agent Research Team – Autonomous research, summarization & report generation
• LLM-as-Judge – LLM-powered evaluation & benchmarking system

• Python, PyTorch, Scikit-learn
• Multimodal RAG, Retrieval Pipelines, Vector Search
• LLM Orchestration, Tool-Use, Multi-Agent Systems
• LLM Evaluation, Guardrails, Observability

• Data Preprocessing, ETL, Clustering
• PostgreSQL, MySQL, pgvector
• Power BI, Reporting & Dashboards

• Node.js, Express, REST APIs
• React, Next.js, JavaScript
• Streaming UIs, Production integrations

• Git, CI/CD, Vercel
• Modular architectures, deployment-first engineering

• Agentic architectures & tool-use patterns
• Multimodal RAG & retrieval pipelines
• Large-scale data clustering & ETL
• LLM evaluation, guardrails & observability
• End-to-end ML deployment on the cloud

Agentic AI • Multi-Agent Orchestration • Multimodal RAG • Retrieval Pipelines • LLM Evaluation • Guardrails • Observability • Clustering • ETL • Next.js • Node.js • pgvector • Vercel

I design systems where agents route, retry, and self-correct with observability baked in — turning messy workflows into production-grade automation 🤖