I build software where research, systems, and product meet.
AI systems · developer tools · native apps · compilers · applied research
I'm a Computer Science & Artificial Intelligence student at the Università della Calabria, an independent software engineer / builder, and the founder of Limes Labs.
I like taking difficult ideas and turning them into working systems. My projects move across AI and agent infrastructure, developer tooling, native software, compilers and formal verification, benchmark systems, local-first products, and applied research.
The part I enjoy most is usually what comes after the first demo: making the system observable, testing where it breaks, measuring the right things, and pushing it until the idea becomes something real.
Default mode: build → measure → break → improve → ship.
|
OpenClaw · upstream contributor Contributions across agent runtime, Gateway reliability and performance, automation, native iOS, and messaging/provider integrations. As of Aug 2026: 23 PRs opened · 18 merged upstream. View upstream contributions · Release-note acknowledgement |
| Project | What I built |
|---|---|
| IntentForm | Local-first design environment where humans and coding agents edit the same validated Semantic Interface Graph, then compile it deterministically to React, Web, Expo and SwiftUI with evidence-bound verification. |
| Sigil | Experimental systems language and compiler with first-class proof obligations, SMT-LIB verification conditions, Z3 integration, memory/ownership models and a native libgccjit path. |
| TracePilot | Reliability studio and eval harness for computer-use agents: step traces, replay, action verification, recovery policies, safety checks, cost accounting and cross-provider evidence. |
| Bite | Native iOS health product with HealthKit, a typed-tool AI coach, long-term memory, encrypted lab-report processing, widgets, Live Activities and a Cloudflare backend. |
| AgentKeys | Native iPhone control surface for coding agents built around structured lifecycle events, capabilities, approvals, sessions and provider-aware controls. |
| Atlas Loop | Runtime-evidence tooling for agents operating real iOS Simulator interfaces: actions, screenshots, app state, logs, network evidence, traces and reproducible handoffs. |
- Elliptic Rank ≥ 30 — certificate-first elliptic-curve research pipeline; independently reproduces the public rank-29 baseline with exact, SageMath and Magma evidence while exploring a constructive route toward rank 30.
- Benchforge — factory for reproducible benchmark arenas with challenge CLIs, independent verification, trust tiers, audit artifacts and hosted leaderboards.
- Ziba — local-first knowledge workspace built around durable Markdown, typed objects, databases, backlinks and graphs.
- TarsGPT — AI robotics runtime with voice, tool use, long-term memory, vision, movement, simulation and gait optimization for a TARS-inspired physical robot.
Languages
Interfaces & product
Infrastructure & data
AI, evaluation & systems
The best way to understand my work is to open a repository and run it.