technical founder + systems engineer
I have spent a large part of my career building startups from the moment when the product is still an argument. At Dreamlit, I rebuilt a prototype into a shipped PS5 and PC title, grew the team from 3 to 50+, raised the company from pre-seed through Series A, and led the platform relationship through certification and launch. I also co-founded Atopia, raised several million dollars, and built the web and VR platform that launched with The Metropolitan Museum of Art.
I always go all the way down the stack when the company needs it. Earlier that meant engine architecture, multiplayer, rendering, and GPU data pipelines. Today much of the work sits close to physical AI: simulation, real-world reconstruction, and the systems that ground models in physical environments. The low-level depth lets me see where the real technical risk lives and build the core before the company can afford a large team.
I work on performance engineering for physical AI systems: simulation, real-world reconstruction, and grounding models in physical environments. That reaches from CPU and GPU kernels and memory movement to inference, rendering, and streaming. The practical goal is simple: run far more simulation and reconstruction work on the same hardware.
A PlayStation-backed, self-published open-world multiplayer game on PS5 and PC. As CTO, I rebuilt the prototype into the shipped game and worked across product, company building, engine architecture, multiplayer, tools, and the foundation systems underneath the world. That included a custom world partition and asynchronous asset and physics streaming in Unreal Engine 4, years before comparable engine support arrived in UE 5.6, 2025. I grew the team from 3 → 50+ across twelve time zones, led Sony certification and milestones end to end, and raised the company from pre-seed through Series A.
I co-founded Atopia and built the company and its web and VR platforms from the ground up. We raised several million dollars to bring renowned museums into curated experiences people could visit from home. The work crossed product, fundraising, museum partnerships, and the engineering needed to launch with The Metropolitan Museum of Art in New York.
Led the architecture behind Netcode for GameObjects, Unity Transport, Multiplayer Tools, Lobby, and Relay, guiding six teams across the multiplayer org, and worked in Unity's performance optimization consulting group reviewing AAA customer codebases, netcode, and hardware performance. Presented the netcode architecture at GameDev Guild 10k+ views.
Across studios and independent projects, I worked on low-level engine, graphics, gameplay, and server systems in C++ and Unreal. Networking remained a particular focus across dedicated-server, listen-server, relay, peer-to-peer, and hybrid topologies. The common thread was production performance: synchronization, replication, transport, platform constraints, and shipping real-time games.
A keyboard-driven pointer system that combines local computer vision, accessibility data, and GPU-rendered hints to find and act on desktop controls. Its C++20 runtime keeps capture, inference, scene publication, interaction, and native input on separate bounded ownership paths behind a versioned C11 ABI.
A C++17 inference runtime where I work on heavily optimized SIMD kernels for quantized flat and graph search, alongside batching, bounded streaming, and the memory systems underneath them.
A self-hosted video intelligence engine that turns MP4, WebRTC, RTSP, and HLS into one ordered event stream. Its bounded Rust media plane separates decode, deterministic frame selection, inference, and durable evidence so an application can work with events without owning the entire video stack.
I still like the part where an idea has to survive contact with a machine. This is one small example from Astral: an 8-bit format used for compact vector search, followed by exact reranking in f32.
five exponent bits, two mantissa bits: range over precision. embeddings stay compact here; the final candidates are reranked in f32.
Long-form notes from the systems underneath the companies and products. Read the writing index.