c++ . low-latency . bhopal, india
I write C++. Mostly the unfashionable parts: epoll loops, order
books, SIMD intrinsics, latency profiles. I contribute to
simdjson (LoongArch LSX and
RISC-V RVV paths for the string builder). I came 3rd at iRage AlgoArena 2026.
Recent CS grad from RGTU Bhopal (May 2026). Most of what I build sits between competitive programming and production systems.
Added LoongArch LSX and RISC-V RVV intrinsics for
find_next_json_quotable_character, following the existing
NEON and SSE2 patterns. Brings vectorised JSON parsing to two more
architectures.
Price-time priority LOB. GoodTillCancel, FillAndKill, Market.
std::map<Price, std::list> for O(log n) insert and O(1)
best-price. Iterator map for O(1) cancel. Fixed a heap-use-after-free
hiding behind a structured binding, collapsed three hash lookups into one,
and ripped an O(n) std::next out of the insert path.
Decoupled trading engine on the Observer pattern with a custom
EventBus. Core-pinned, cache-warmed,
-march=native. 500k events/sec, p95 under 150µs.
Live Binance order books over async SSL WebSocket (Boost.Beast), parsed
with simdjson. Unit and integration tests on GoogleTest.
Single-threaded, edge-triggered epoll RESP server. Against
Valkey 8.0.7: +11% on pipelined writes (1.22M req/s) and 3× lower
p99 latency. Fixed a partial-send() desync with a per-client
write buffer behind EPOLLOUT, killed a wedged 8 KB read
buffer (now dynamic, 64 MB cap), and cut epoll_wait
syscalls 91% with batch tuning. Clean shutdown over signalfd.