M

meihua_7

系统 · 全栈工程师 · Solana 与量化交易 Systems · Full-Stack Engineer · Solana & Quant Trading

端到端构建加密交易系统——链上实时数据、低延迟执行、套利与做市，以及配套风控。主力 Rust 与 Python，大多独立设计与交付。 I build crypto trading systems end to end — real-time on-chain data, low-latency execution, arbitrage and market making, with the risk controls around them. Mostly Rust and Python, and largely designed and built solo.

𝕏 @meihua_7 🔗 prios.xyz
🎒 申请方向Applying for · Systems Engineer (Rust) / Full-Stack Engineer @ Backpack
关于我About

系统与量化工程师，专注加密交易基础设施，横跨 Solana DeFi、预测市场与中心化交易所。用 Rust 编写高性能链上与流式组件——原生 Solana 程序、链上原子套利、闭源 AMM 逆向、Yellowstone gRPC 实时数据、MEV 竞速上链；用 Python 构建策略、数据与风控管道。熟悉 Solana RPC 节点的部署运维与端点延迟优化。从后端一直到产品与前端都能独立完成——prios.xyz 系列工具的设计与前端也出自我手。这些系统大多由我独立设计与交付，并深度结合 AI 编程工具高效产出。我也是 Backpack 的实际用户，围绕它构建过自己的工具——对这个平台不是旁观者。对交易所内部机制、订单簿与市场微观结构有强烈兴趣。 Systems and quantitative engineer working on crypto trading infrastructure across Solana DeFi, prediction markets, and centralized exchanges. In Rust I build high-performance on-chain and streaming components — native Solana programs, on-chain atomic arbitrage, closed-source AMM reverse engineering, real-time Yellowstone gRPC data, and latency-competitive MEV submission; in Python I build strategy, data, and risk pipelines. Comfortable deploying and operating Solana RPC nodes and tuning endpoint latency. I also take the work all the way through to product and frontend — the prios.xyz tools were designed and built by me. Most of these systems were designed and shipped solo, with heavy and deliberate use of AI coding tools. I'm also an actual Backpack user who has built tooling around it — not a bystander to the platform. Deeply interested in exchange internals, order books, and market microstructure.

技能Skills

Languages

RustPython TypeScriptSQL

Blockchain & Trading

Solana native programsCPI Versioned TxALT Yellowstone gRPC / GeyserRPC node deployment Jito & MEV relaysJupiter AMM Raydium / Orca / MeteoraPolymarket KalshiCEX APIs Cross-venue arbitrageMarket making Black-ScholesKelly sizing

Backend, Data & Web

Tokiotonic / gRPC asyncio / aiohttpWebSocket order books FastAPINext.js / React PostgreSQLClickHouse Redis / StreamsDashMap Borsh / zero-copyDocker

Practices

AI-assisted dev (Claude Code)Systems thinking Product sense & tasteEvent-driven / microservices Real-time streamingRisk & position limits Backtesting / paper tradingOn-chain reverse engineering
精选项目Selected Projects
fluxarb★ 技术含金量最高★ most technical Rust

Solana 链上原子套利 botSolana on-chain atomic arbitrage bot

  • 构建 Rust workspace（链上程序 + 链下 bot），在单笔 Solana 交易内原子执行两腿 DEX 套利，未通过最小利润检查即在链上回滚。Built a Rust workspace (on-chain program plus off-chain bot) that executes two-leg DEX arbitrage atomically within a single Solana transaction, reverting on-chain when a minimum-profit check fails.
  • 编写原生 Solana 程序（非 Anchor），手写指令编码，跨 30+ 个 DEX 做 CPI 集成，覆盖 Raydium、Orca、Meteora、Phoenix、OpenBook、Pump.fun。Wrote a native Solana program (no Anchor) with hand-encoded instructions and CPIs into 30+ DEXs — Raydium, Orca, Meteora, Phoenix, OpenBook, Pump.fun.
  • 逆向工程三个闭源、无文档的 AMM（Obric V2、SolFi V1 与 V2）：通过解码链上交易还原账户布局、指令判别器与数据偏移。Reverse-engineered three closed-source, undocumented AMMs (Obric V2, SolFi V1/V2) by decoding on-chain transactions to recover their account layouts, instruction discriminators, and byte offsets.
  • 执行链路经 Yellowstone gRPC 订阅程序账户更新、模拟利润，借助 ALT 与动态计算预算，并发通过 Jito bundle 与多家 MEV 中继提交。Built an execution path that subscribes to program-account updates over Yellowstone gRPC, simulates profitability, and submits concurrently through Jito bundles and multiple MEV relays using Address Lookup Tables and dynamic compute budgets.
pred-arb Python

预测市场跨场所套利与做市系统Prediction-market cross-venue arbitrage & market making

  • 构建异步系统，并发追踪 Polymarket、Kalshi、Opinion 的订单簿，在二元与多结果市场上检测并执行跨场所套利。Built an async engine that tracks order books on Polymarket, Kalshi, and Opinion concurrently to detect and execute cross-venue arbitrage on binary and multi-outcome markets.
  • 实现无风险双边策略：当 YES 与 NO 的合计卖价扣费后低于面值时同时买入两边，结算后合并仓位赎回抵押品。Implemented a risk-free two-sided play: buy YES and NO together when their combined ask trades below par net of fees, then merge the positions to redeem collateral.
  • 开发由 Binance 实时价格与波动率驱动的 Black-Scholes 定价引擎，发现短周期加密市场的定价偏差，并做预言机滞后修正与临近结算停手。Built a Black-Scholes pricing engine fed by real-time Binance price and volatility to surface mispricings on short-horizon crypto markets, with oracle-lag correction and an end-of-window cutoff.
  • 构建体育做市模块：从多个赔率源（Bet365、Betfair、Pinnacle）推导公允价，在 Polymarket 双边挂单，并施加分层的敞口与频率限制。Built a sports market-making module that derives fair value from several odds sources (Bet365, Betfair, Pinnacle) and quotes two-sided on Polymarket under layered exposure and frequency limits.
solana-trading-bot Rust + Python

实时 Solana 交易 botReal-time Solana trading bot

  • 设计 Rust + Python 混合微服务：Rust 采集端经 Yellowstone gRPC / Geyser 摄取链上数据，Python 分析端负责打分、决策与执行。Designed a hybrid Rust + Python system: a Rust collector ingests on-chain data over Yellowstone gRPC / Geyser, and a Python analyzer handles scoring, decisioning, and execution.
  • 构建 Pump.fun 指令解析器，以 Anchor 判别器识别 create / buy / sell，并用 Borsh 解析 bonding curve 状态。Built a Pump.fun instruction parser that identifies create/buy/sell by Anchor discriminator and decodes bonding-curve state with Borsh.
  • 用 solders 构造 versioned transaction，支持可配置优先费与 Jito bundle 提交，含 dry-run 与实盘双模式。Implemented execution with solders-built versioned transactions, configurable priority fees, and Jito bundle submission, across dry-run and live modes.
  • 搭建多存储数据层（ClickHouse 时序、PostgreSQL 状态、Redis Streams 消息），Docker Compose 编排，含代币生命周期状态机与 rug 黑名单。Architected a multi-store data layer — ClickHouse for time series, PostgreSQL for state, Redis Streams for messaging — orchestrated with Docker Compose, plus a token-lifecycle state machine and a rug blacklist.
info-arb Python

信息驱动的预测市场套利（多 Agent）Information-driven prediction-market arbitrage (multi-agent)

  • 构建多 Agent 系统（feed、model、risk、execution、orchestrator），摄取实时新闻与宏观日程事件，在 Polymarket 上交易由此产生的定价偏差。Built a multi-agent system (feed, model, risk, execution, orchestrator) that ingests real-time news and macro-calendar events and trades the resulting mispricings on Polymarket.
  • 融合多个 LLM 的概率估计，做加权、一致性惩罚与 Platt 校准，仅当模型与市场的 edge 超过阈值时才出手。Fused probability estimates from several LLMs with weighting, consistency penalties, and Platt calibration, acting only when the model-vs-market edge clears a threshold.
  • 经 WebSocket 订阅 Polymarket CLOB 订单簿，施加单笔 / 单市场上限、亏损限制、冷却与回撤熔断。Subscribed to Polymarket CLOB order books over WebSocket, with per-trade and per-market caps, loss limits, cooldowns, and a drawdown circuit breaker.
prios · prios.xyz Python + Next.js

可验证的量化研究与纸上交易系统（面向社区）Verifiable quant research & paper-trading system (community)

  • 构建面向社区的免费系统，横跨预测市场与加密，组织为多个策略"经理"（规则套利、新闻冲击、上新、聪明钱跟踪、CEX-DEX 结构），在统一风控下运行。Built a free, community-facing system spanning prediction markets and crypto, organized as strategy "managers" (rule arbitrage, news shocks, listings, smart-money tracking, CEX-DEX structure) under one risk-governance layer.
  • 要求每条信号记录来源、价格、可成交流动性与结算规则，写入 append-only 的 JSONL 账本，实现完整审计与回放。Made every signal verifiable — source, price, executable liquidity, and settlement rules recorded in append-only JSONL ledgers for full audit and replay.
  • 独立完成从策略后端到 prios.xyz 前端的全部实现，并公开发布给社区使用。Owned everything from the strategy backend to the prios.xyz frontend, and shipped it publicly for the community.
开源与社区Open Source
grpc-benchmark-ts TypeScript · OSS

为社区编写的 Solana gRPC / Yellowstone 端点延迟对比工具，已被社区采用。它以"同一 slot 在多个端点的到达先后"衡量真实场景下的相对延迟，并提供串行与并发两种 ping-pong 延迟测试。对做 MEV、抢跑与实时链上应用的团队来说，端点选型直接决定竞速成败——这个工具让选型有据可依，也是我把节点与基础设施经验回馈社区的方式。 An open-source tool for benchmarking and comparing Solana gRPC / Yellowstone endpoint latency, adopted by the community. It gauges real-world relative latency by which endpoint sees the same slot first, alongside serial and concurrent ping-pong tests. For anyone doing MEV, front-running, or real-time on-chain work, endpoint choice decides the race — this makes that choice data-driven, and is how I give my node and infrastructure experience back to the community.

在 GitHub 查看 →View on GitHub →
工作方式How I Work

到 2026 年，写代码本身正在被 Codex、Claude Code 这类工具重新定义。把一个功能实现出来，已经不再是瓶颈——真正拉开差距的，是全局的系统思考、驾驭 AI 的能力，以及对"做什么、不做什么、什么才算好"的判断力与品味。 By 2026, writing code itself is being redefined by tools like Codex and Claude Code. Shipping a feature is no longer the bottleneck — what sets people apart now is systems-level thinking, the ability to direct AI well, and the judgment and taste to know what to build, what to skip, and what "good" actually looks like.

上面这些系统，几乎都由我一个人设计并落地：我来定义问题、划定架构与边界、把控取舍，把大量执行交给 AI。这也是我不愿把自己框进单一岗位的原因——从链上底层，到产品与前端形态，再到回馈社区的工具，我在意的始终是把正确的事做出来，并且做得漂亮。 The systems above were almost all designed and shipped by one person: I define the problem, set the architecture and boundaries, own the trade-offs, and hand most of the execution to AI. That's why I don't box myself into a single role — from on-chain internals to product and frontend to community tooling, what matters to me is building the right thing, and building it well.

为什么是 BackpackWhy Backpack

Backpack 由 Armani Ferrante 创立——他是 Anchor 的作者，而 Anchor 是 Solana 开发的事实标准框架；他也是 Mad Lads 和 Backpack 钱包背后的人。换句话说，这是一家由 Solana 生态核心开发者创办的交易所。对一个 Solana 开发者来说，这样的团队很有吸引力。 Backpack was founded by Armani Ferrante — the creator of Anchor, the de facto standard framework for Solana development, and the person behind Mad Lads and the Backpack wallet. In other words, it's an exchange built by a core Solana developer. For a Solana developer, that's a genuinely compelling team to join.

而我自己也是 Solana 生态的开发者：写链上原子套利程序、逆向闭源 AMM、跑 Yellowstone gRPC 实时数据、部署与运维 RPC 节点，并把端点延迟基准工具开源回馈社区。我也是 Backpack 的实际用户，围绕它构建过自己的工具——对这个平台不是旁观者。 I'm a Solana developer myself: I write on-chain atomic arbitrage programs, reverse-engineer closed-source AMMs, run real-time Yellowstone gRPC data, deploy and operate RPC nodes, and open-sourced an endpoint-latency benchmark for the community. I'm also an actual Backpack user who has built tooling around it — not a bystander to the platform.

我是真心喜欢 Solana 这个生态，也愿意为它长期投入、认真做贡献。Backpack 把一流工程与 Solana 原生体验结合的方向，正是我想为之努力的事。 I genuinely care about the Solana ecosystem and want to invest in it for the long run. Backpack's direction — pairing world-class engineering with a Solana-native experience — is exactly what I want to work on.