Wade Spires

Staff engineer at Snap. I build low-latency C++ systems — messaging infrastructure, electronic trading, LLM inference routing — and ship my own developer tools and iOS apps.

About

I write C++ for systems where latency and correctness both matter. At Snap I architected the cross-platform notification infrastructure shared by the Snapchat iOS and Android apps. Before that I built direct market-access trading infrastructure at Clover Light Capital, and low-latency trading systems at Automated Trading Desk and Citi.

The through-line is concurrent state made simple enough to reason about: single-threaded cores behind async façades, shared-nothing designs, no locks on the hot path. Lately that has meant the same problems one layer up: Ranvier routes LLM inference requests to the backend that already holds the relevant KV cache, and pare trims the command output an AI agent reads down to the lines that matter.

Outside work I build iOS apps end to end, from the first line of code to the App Store listing: Gym Log+ since 2013, and Rowdy Rows, a game released this year.

Projects

Systems

Ranvier

2025 – presentC++20, Seastar

An open-source router for self-hosted LLM fleets. It sends each request to the GPU that already holds its prompt prefix in KV cache, rather than to whichever backend is free, so the prefill is not computed twice. Routing is an adaptive radix tree over token prefixes, with O(L) lookups under 50µs independent of how many prefixes are cached, in a shared-nothing, thread-per-core design on Seastar with no locks on the hot path. Circuit-breaking health checks, clustering over encrypted gossip, and Kubernetes discovery with a Helm chart.

  • 3–5×KV-cache hit rate
  • 27–29%lower median TTFT
  • 8× A100measured on

ranvier.systemsGitHub

pare

2026Rust

A Unix filter that cuts noisy command output (test runs, builds, logs) down to a token budget for AI agents, without losing anything. It keeps the lines that differ, such as the failing test, its stack trace and the verdict, counts the shapes that repeat, and saves the raw input so any omitted line can be pulled back on demand. Deterministic, so the same input always gives the same summary. Open source, released October 2026 as a single native binary.

  • 60×smaller output
  • 0failures lost, blind test

github.com/wspires/pare

Apps

Gym Log+

2013 – presentSwift, SwiftUI

A strength training app for iPhone and Apple Watch, built and shipped solo: roughly 450,000 downloads. In 2026 I rewrote it from scratch, from Objective-C to SwiftUI, and shipped the rewrite as an update to the same App Store listing, so every existing user’s history had to import losslessly from formats the old app had written for over a decade. CloudKit sync, a standalone watch app, split-view layouts for iPad and the foldable iPhone Duo, and no third-party dependencies.

gymlogplus.com

Rowdy Rows

2026Swift, SwiftUI, Metal

An iPhone game about an inbox that won’t stay cleared. One spring simulation drives both the animation and the haptics, so a row’s weight can be told by feel alone. Every level is playtested by a simulated player: its runs set the star times, and a slower version checks that no level is unwinnable for an average player.

rowdyrows.com

Experience

  1. – Present

    Staff Software Engineer Snap Inc.

    Primary C++ engineer on Snapchat's cross-platform notification infrastructure, shared between Android and iOS.

    Owned its concurrency model, persistence, and delivery-state handling.

    Technical lead on Streak Restore with Snapchat+, and lead of 5–10 engineers across backend, iOS, and Android on messaging features including Chat Reply and Chat Mentions.

  2. –

    Software Engineer Clover Light Capital

    Built proprietary futures trading infrastructure in modern C++ for CME, CFE, and Eurex. Replaced a vendor Eurex implementation with direct EOBI market data and ETI order entry, cutting median per-message latency by 55 microseconds and eliminating the licensing cost, and wrote the direct handlers for CME MDP 3.0, CFE PITCH, and CFE BOE.

  3. –

    Software Engineer (AVP from 2013) Automated Trading Desk / Citi

    Developed, deployed, and operated low-latency trading systems in C++, Python, Perl, and Bash alongside teams across EMEA, Japan, and Australia. Moved order routing for BATS and Chi-X in EMEA to direct market access, cutting order latency from milliseconds to microseconds. Extended the C++ market data readers with full order book depth and auction-phase indicative pricing, and moved SBI Japannext market data off a vendor feed onto the exchange’s native ITCH protocol.

Education

  • M.S. Computer Science University of Central Florida

  • B.S. Computer Science, B.S. Mathematics University of Tennessee

Elsewhere