Software Engineer · Zurich, Switzerland
Distributed systems. Data infrastructure.
AI‑augmented engineering.

I'm a software engineer at UBS in Zurich, currently building the model-routing service for the bank's internal AI platform — after building data platforms and LLM agent tooling for its engineering workflow. Before UBS: hunting hypervisor bugs at HPE, and a master's thesis in the US building a C++ distributed protocol for drone swarms — consensus, cryptographic integrity, and hard numbers to prove it works.
I like systems you can measure: latency, throughput, fault tolerance. And I like shipping — from first design to monitored production deployment.
“When coding is magic.”
Building the model-routing service for the bank's internal AI platform — directing each prompt to the best-suited model across self-hosted vLLM endpoints and Azure AI Foundry, running on Kubernetes. Provisioning the supporting Azure infrastructure as code.
Part of UBS's AI-augmented engineering practice: wiring MCP tool definitions into agent harnesses so LLM agents could assist with code analysis and review. Authored multithreaded test suites, wired in as release gates, catching concurrency regressions before they reached production.
Designed Python ETL pipelines correlating 2M+ rows from 300+ production apps across EMEA into real-time dashboards — cutting manual triage by 35%. Deployed through CI-gated pipelines with RBAC and audit logging.
Isolated 20+ critical defects in HPE VM Essentials pre-production — root-causing through Linux log analysis and KVM/hypervisor cross-checks, automating failure reproduction with Python, Bash, Docker and Kubernetes.
Built SAHARA: a C++ data-dissemination architecture for infrastructure-less drone swarms — reputation-based consensus, SHA-256 block-chaining, Bloom-filter reconciliation. Validated over 500+ automated NS-3 campaigns up to 196 nodes.
Master's thesis
196 drones. No infrastructure. Every node must trust the data it receives — so SAHARA chains it with SHA-256, reconciles it with Bloom filters, and votes on it with reputation-based consensus.
Personal project · 2026
A copy-trading engine for Polymarket (Polygon/EVM). It watches target wallets over a WebSocket stream, reconciles against the Data API, and routes FAK orders through the CLOB — with slippage caps, per-trader limits, crash-safe state, and AES-256-GCM-encrypted credentials.
Deployed on AWS EC2 behind a GitHub Actions CI/CD gate. Ships its own paper-trading Test Lab simulating market impact, latency, and partial fills.
C++ distributed data dissemination for drone swarms — consensus, SHA-256 chaining, Bloom filters. 500+ simulation campaigns.
Real-time on-chain copy-trading engine for Polymarket — WSS ingestion, FAK execution, full risk layer. 4.5K+ LOC.
Job-market intelligence webapp — 7 ATS integrations, 24 companies, explainable CV-fit scoring, optional local-LLM matching.
Full-stack geospatial platform — 80K+ LOC. Swift/CoreML client over a Python/GDAL pipeline, sub-100ms p99 tile latency on AWS.
C++ operator-graph scheduling solver for the MLSys 2026 competition, produced by an agent-driven development loop I designed and ran — I owned the architecture, the evaluation contract and the submission review.
Applying blockchain structures to infrastructure-less mesh networks — the research seed that grew into SAHARA.
Gas-efficient on-chain key-value store built on a Red-Black Tree, exploring EVM storage-cost trade-offs.
Fully on-chain Battleship on Solana — game logic and state as a Rust smart contract built with the Anchor framework.
8× speedup on a 16-node cluster for mazes up to 64M pixels — distributed search with OpenMPI.
Digging into execution clients: evmone AOT-compiler and go-ethereum — understanding the EVM below the bytecode.
Native iOS travel app — SwiftUI, clean architecture, Apple Developer Program member.
Politecnico di Milano · 2022 — 2024
110/110 cum laude
University of Pavia · 2018 — 2021
98/110
Zurich, Switzerland · open to interesting problems.