1. 📈 hft-limit-order-book (Modern C++20 HFT Matching Engine)
A sub-microsecond Limit Order Book engine designed for deterministic, sub-microsecond execution on Unix platforms.
- Core Technology Stack: C++20, CMake, Ninja, Google Benchmark.
- Low-Latency Architecture:
- Memory Pool: Zero runtime heap allocation on the matching path using
MemoryPool<T>block allocators. - Ring Buffer: Thread-safe order intake via lock-free Single-Producer Single-Consumer (SPSC) atomic queues.
- Memory Alignment: Aligned structures (
alignas(64)) to completely prevent false-sharing on cache lines.
- Memory Pool: Zero runtime heap allocation on the matching path using
- Latency Profile:
- Mean: 215.42 ns
- p99 (Taker fill): 312.00 ns
- p99.9 (Extreme tail): 480.00 ns
2. ⚡ agentic-event-orchestrator (High-Throughput Asynchronous Agent Bus)
A highly concurrent event dispatcher and LLM router built in Python utilizing uvloop and Redis Streams.
- Core Technology Stack: Python 3.10+, uvloop, FastAPI, Redis Streams, Pytest.
- Performance Metrics:
- Throughput: 18,000+ requests/sec on single-threaded execution loops.
- Routing Overhead: p99 < 8.5ms routing time for multi-agent workflows.
- Resilience: Embedded memory queue failover in case of Redis stream drop-outs.
- LeetCode Guardian: Handle eayushsingh (Contest Rating: 2200+ | Top 0.5% globally).
- Codeforces Candidate Master: Handle
eayushsingh(Peak Contest Rating: 1900+). - Open Source Contributions: Contributor to Supabase (UI layout patches) and Apache Fory (compiler integration, Maven gRPC generators).
| Category | Skills & Technology Stack |
|---|---|
| Low-Level Systems | C++20 · Rust · Linux Kernel Tuning · CPU Pinning · Cache Optimization |
| High-Throughput Pipelines | Redis Streams · gRPC · ZeroMQ · WebSockets · uvloop · Multi-Threading |
| Data & Core Platforms | PostgreSQL · Supabase · Redis · Docker · Spring Boot · Next.js · FastAPI |
| Tooling & Profiling | CMake · Ninja · valgrind · perf · GDB · AddressSanitizer · ThreadSanitizer |
sequenceDiagram
participant TickSrc as Exchange Feed (Ingestion Thread)
participant RingBuf as Lock-Free SPSC RingBuffer
participant MatchingEngine as LOB Matching Engine (Core Pin Thread)
participant MemoryPool as pre-allocated Memory Pools
participant OrderBookMap as O(1) direct Order Map
participant TradesOut as Trade Execution Feed
TickSrc->>RingBuf: atomic enqueue (Side, Qty, Price)
RingBuf->>MatchingEngine: atomic dequeue
MatchingEngine->>MemoryPool: allocate Order/Limit node (O(1))
MatchingEngine->>OrderBookMap: index order pointer (O(1))
MatchingEngine->>TradesOut: execute matching trades (sub-microsecond)