MLH × Meta — Production Engineering Fellowship
Production Engineering Fellow · cohort 26.SUM.B
A 12-week fellowship mentored by Meta engineers, focused on the operational side of running software in production.
View →
About
I'm a student at Wesleyan, and this summer I'm a Production Engineering Fellow with MLH and Meta.
Most of what I build sits low in the stack. The project I've sunk the most hours into is a limit order book in OCaml that I got down to zero allocations on the matching path. There's also a Rust tool that rebuilds packages and checks them byte for byte, and a small Go proxy that sits between an LLM and its tools to screen what it's allowed to call. Lately I've been benchmarking the same workloads across a few Linux distros on identical VMs, just to see what actually differs.
The thread through all of it is that I'd rather measure something than take it on faith. That's most of why production engineering appealed to me.
Experience & projects
Enclave
Local-first clinical-document pipeline
Agentic extraction + ICD-10/CPT matching with PHI egress (bytes off-machine) tracked as a first-class metric, measured across a rules floor, an on-device 3B, and hosted baselines.
PHI egress = 0 B on-device path
View →
ClearHash
Rust supply-chain integrity verifier
Rebuild every package, compare every byte, block every tamper. Reproducible-build verification driven by Sigstore + SLSA.
byte-for-byte rebuild diffing
View →
Halberd
JSON-RPC firewall for MCP agents
Inspects every tools/call between an LLM and its MCP servers, blocking argument injection and capability creep before they reach the host.
inline call inspection
View →
Quarry
Bare-metal MEV arbitrage bot
TypeScript mempool scanner + Yul executor + Aave V3 flashloans. Built for the smallest possible runtime and gas footprint.
188 B runtime · 110k gas/2-hop · 99.89% acc.
View →
ocaml_limit
Zero-allocation Limit Order Book engine
High-performance LOB matching engine in OCaml 5 with a real-time, high-density dashboard inspired by the Bloomberg Terminal.
0 allocations per submit
View →
CapitolAlpha
Wesleyan QAC 420 — quantitative finance
Pipelined 16,203 disclosed Congressional stock trades (2020–2024) and measured a +2.58% Jensen's alpha vs the S&P 500, significant at p < 0.05.
+2.58% α · p < 0.05 · n = 16,203
View →
LinuxBenchHub
VM benchmarking platform
Runs the Phoronix Test Suite across Ubuntu, Fedora, and Debian under identical virtual hardware, with a Rails 8 dashboard and an embedded noVNC live view.
identical-hardware cross-distro runs
View →
Education
Wesleyan University
B.A. — Computer Science & Economics (double major)
Middletown, CT · Coursework through the Quantitative Analysis Center (QAC): built CapitolAlpha for QAC 420 and ran the DuckDB pipeline for DataFest 2026.
Dominican University of California
ACS-4996 — Supervised Work Experience (MLH × Meta Fellowship)
Remote · Summer 2026 · Concurrent enrollment carrying the MLH Production Engineering Fellowship as for-credit work experience (10 units).
Places I've been
2 places on the map.