Hi, I'm Filippo

AI Software Engineer. I build multi-agent systems and AI pipelines. Open source author, published researcher and hackathon enjoyer. 🤖🚀

About Me

Hey! I'm Filippo, currently working as an AI Software Engineer at Blue Reply, where I build agentic AI and multi-agent systems. Before, I worked on backend infrastructure for autonomous vehicles at scale. In my spare time I maintain CtxVault, an open source memory layer for AI agents.

Work Experience

Blue Reply IT

October 2024 - Present

AI Software Engineer

• Designed a multi-agent system (LangGraph, Python) for automated end-to-end JUnit test generation from GitLab issues. Adopted by 2+ engineering teams within 3 months, saving ~60 QA hours/quarter and fully replacing manual workflows.
• Fine-tuned an LLM (Qwen) via SFT+DPO for Text-to-SQL generation on a complex automotive schema (~50 tables), reaching 70% execution accuracy vs. a 30% baseline — deployed in production for internal data querying.
• Built a real-time clustering system (Spring Boot, K8s, AWS, MongoDB) for road events powering autonomous driving features across 1M+ vehicles — 85% latency reduction, scaling to 30k+ daily events.

CtxVault

December 2025 - Present

Open Source Author & Maintainer

• Built a local memory infrastructure for AI agents — typed vaults (semantic + procedural), structural isolation, and persistent cross-session retrieval via vector search (FastAPI, ChromaDB, Python).
• Shipped to PyPI with MCP server support out of the box; 3.5k+ total downloads, 35+ GitHub stars in the first week.

SPOT Software

February 2022 - April 2022

Software Engineer, Intern

• Developed modular web applications using Angular microfrontends, enabling independent deployment pipelines across feature teams.
• Implemented async data flows with RxJS streams and leveraged functional programming patterns (immutability, pure functions, monadic composition) to reduce side effects and improve code testability.

Vae Victis Games

January 2018 - March 2018

AI Tester, Intern

• Evaluated and optimized the AI model for random circuit generation in the sandbox racing game RaceCraft, enhancing track diversity and player experience. Tested and fine-tuned hyperparameters for AI-driven automated drivers, resulting in smoother, more competitive, and adaptive racing behavior.

My Education

University of Bologna

2022 - 2024

Msc in Computer Science and Engineering

• 110/110 with Honors
• Thesis focused on decentralized multi-agent reinforcement learning via GNNs, leading to a published research paper.

UPC Barcelona

September 2023 - February 2024

Erasmus - MSc in Artificial Intelligence

• Completed coursework in Computer Graphics, Virtual and Augmented Reality, Cognitive Interaction with Robots, and Normative and Dynamic Virtual Worlds, focused on real-time simulation, embodied interaction, and AI for games.
• Simulated particle, cloth, and fluid dynamics in C++ for real-time rendering.
• Built AR and VR apps in Unity, including mobile-based AR and room-scale VR, sustaining 60 FPS on consumer devices.
• Engineered a 3D investigation game with FSM-driven NPCs and AI behaviors, enabling 100% dynamic branching dialogue across 5+ suspect profiles.
• Developed a pose-matching browser game using Google MediaPipe and webcam-based real-time human pose estimation, achieving >90% classification accuracy and <100 ms latency on Chrome.

University of Bologna

2019 - 2022

Bsc in Computer Science and Engineering

• 110/110 with Honors
• Thesis: built a real-time network architecture in Unity (C#) for collaborative Mixed Reality on HoloLens 2, supporting consistent multi-user state sync. Achieved <50 ms latency and 3,800 packets/sec throughput in high-entity-density scenarios over TCP and UDP.