Ran (Eddie) Dang

Staff Software Engineer — AI Infrastructure & Developer Tools
|

Email LinkedIn GitHub Resume.pdf

Singapore · Open to Singapore & North America

0Yrs Engineering
0Open Source
0Agent Runtimes
0Users Served
Staff Engineer / Tech Leader @ OKX 2024.07 - Now
Infrastructure & Reliability Engineering · Singapore
  • Autonomous Agent Operations: Before OpenAI's Symphony was open-sourced, independently built Lindy Orchestrator grounded in harness engineering theory. After Symphony's release, studied its architecture deeply and integrated enterprise-grade capabilities. Now adopted by 4–5 engineering teams with agents working near-autonomously around the clock, zero human intervention, quality controlled via automated QA gates.
  • Company-wide JVM Platform Migration: Led migration of 800+ Java services from ZingJDK (C4GC) to OpenJDK 21 (Generational ZGC). Built custom JVM benchmark framework. GC pause reduced from tens-of-ms to <1ms, saving annual licensing costs.
  • 0-to-1 SLO Platform: Designed complete SLO system—data pipelines, metrics, alerting, dashboards—reliability backbone for 1,000+ microservices.
  • Automated Incident Response: Topology-aware anomaly detection and 5-minute RCA system across billions in daily trading volume.
  • High-Performance Routing Engine: Optimized routing engine integrating 300+ external protocol APIs and 5M+ data points with sub-second latency.
Staff Engineer @ TikTok 2023.01 - 2024.07
E-Commerce Content Trust & Safety · Singapore
  • Global Platform Scale: Strategic Tech Lead for Content Moderation Platform, enabling safe global expansion of TikTok Shop with multi-country regulatory compliance.
  • Architectural Agility: Engineered event-driven governance strategy platform from 0-to-1, providing flexibility to launch new markets with minimal engineering overhead.
  • Data-Driven Strategy: Championed a parallel risk measurement system providing leadership with real-time visibility into content risk levels and informing key business decisions.
  • Team Culture: Established the team's engineering culture from the ground up — development specifications and quality gates sustaining high-velocity delivery.
Earlier Experience 2016 - 2023
  • Meituan (2021–2023): Led operational infrastructure digital transformation for Mobike.
  • YuanFuDao (2019–2021): Scaled infra to 20K concurrent users. RFC-driven technical strategy.
  • Meituan (2016–2019): Local services platform engineering.

Skills

Languages

PythonGoJava (JVM internals) TypeScriptRust

Backend & Data

FastAPIKafkaFlink PostgreSQLRedisElasticSearch

Infrastructure

K8sDockerAWS GCPPrometheusGrafana InfluxDBOpenTelemetry

AI Engineering

Claude (MCP, Agent SDK)Multi-Agent Orchestration LLM Tool UsePrompt Engineering Harness Engineering

Other

Bilingual (EN/CN)Cross-functional Collab API Design

Education

Beijing Institute of Technology 2012 - 2016
B.S. Internet of Things, Computing Academy
skillpmUniversal Package Manager for AI Agent Skills
Open Source
GitHub

A CLI that installs, manages, and secures skill packages across 10+ AI agent runtimes. Think npm for AI agent capabilities — one command to deploy a skill to Claude Code, Codex, Gemini CLI, Copilot, Cursor, and more.

Go Cobra CLI TOML Homebrew 10+ Runtimes
Architecture
skillpm CLI install / sync / scan Registry versions / metadata Security injection / entropy AGENT RUNTIMES Claude Code Codex Gemini CLI Copilot Cursor TRAE OpenCode Kiro

> Universal Install

One CLI to deploy skills to any supported agent runtime with version control and lockfiles.

> Security Scanning

Built-in prompt injection detection, entropy analysis, and dangerous pattern matching.

> Project Manifests

TOML-based manifests with lockfiles for reproducible AI development environments.

> Zero Cloud

Entirely local. No accounts, no telemetry, no cloud dependencies.

lindy-orchestratorMulti-Agent Work Harness
Published on PyPI
GitHub PyPI

A "harness engineer" for AI coding agents. Decomposes goals into DAG-scheduled tasks, dispatches to isolated git worktrees, validates via a 6-stage QA pipeline, and teaches agents HOW to fix failures through structured remediation — not just WHAT failed. Repo is the single source of truth: zero database, pure git.

Python 3.11+ Typer CLI Pydantic Rich UI Git Worktrees PyPI GitHub Actions DAG Scheduler
lindy-orchestrator — terminal
Lindy Orchestrator in action
Architecture — v0.13.0
INPUT Goal NL / file GitHub Issue gh CLI fetch Resume checkpoint PLANNING Planner reads STATUS.md + ARCHITECTURE.md → LLM → TaskPlan (JSON DAG) Context Engine STATUS.md per module ARCHITECTURE.md CONTRACTS.md EXECUTION Scheduler DAG topo-sort · max 3 parallel per-module concurrency limits lifecycle hooks · checkpoints Git Worktrees .worktrees/task-{id}/ Dispatcher JSONL streaming events stall: warn@150s kill@600s Claude Code Codex CLI Prompt Injection CLAUDE.md + STATUS.md + mailbox boundaries + negative constraints VALIDATION CI Structural Command Layer Agent Custom QA Gate Pipeline (sequential) Feedback pytest / ruff / tsc parsers progressive retry focus FAIL retry (max 2x) STATE & CONTEXT — zero database, pure git STATUS.md per-module health surgical updates CLAUDE.md root + per-module agent instructions Mailbox JSONL inter-module thread-safe append Sessions JSON checkpoints resume support Logs JSONL audit trail append-only ANALYSIS Entropy Scanner Live Dashboard Garbage Collector Discovery System

> Worktree Isolation

Each parallel task in its own git worktree. No checkout races, full isolation, thread-safe.

> Negative Boundaries

ARCHITECTURE.md enforces what agents can't do — more powerful than positive constraints.

> Structured Remediation

QA failures parsed by gate-specific analyzers (pytest, ruff, tsc) into actionable fix guidance.

> Zero Database

Pure git + markdown + JSONL. Fully auditable, works offline, no deployment needed.

> Provider Abstraction

Pluggable dispatch protocol. Swap Claude Code ↔ Codex per module without changing orchestration.

> Hot-Reload Config

Safe config sections (safety, QA gates, hooks) reload mid-execution for live adjustments.

lindyAI-Coordinated Quant Trading System
Testnet Ready

4 autonomous departments (Engineer, Research, Data, QA) each managed by AI agents. Testnet-ready with Funding Carry V3 strategy: Sharpe 4.9, 10 symbols, delta-neutral.

Rust TypeScript Python Tauri Parquet 3 Exchanges 12.8M Rows
Lindy Trading — Tauri
Lindy Strategies Lindy Trading Lindy Data
Strategies
Multi-Department Architecture
CONTRACTS.md cross-department interfaces Engineer Rust / 8 crates / Tauri Research Python / quant modeling Data 12.8M rows / 3 exchanges QA validation / tests Funding Carry V3 Sharpe 4.9 / 10 symbols / delta-neutral

> Multi-Department AI

4 autonomous departments each managed by AI agents, coordinating through shared contracts.

> Rust Engine

8-crate Rust monorepo for the trading engine with Tauri desktop UI.

> 12.8M Data Points

Parquet pipeline ingesting from 3 exchanges. Full dataset backtesting before promotion.

> Cross-Validation

0.000 diff in Sharpe between Research and QA. 1,396 trade verification.

roxyOptions Portfolio Manager
Private

US equities options portfolio manager with strategy scanner. Covered calls, cash-secured puts, protective puts, collars. MooMoo-style options chain UI.

Python FastAPI SQLite Playwright GitHub Actions
Roxy - Options Portfolio
Dashboard Strategy Lab Options Chain Positions
Dashboard

> Strategy Scanner

Scans for optimal covered call, cash-secured put, protective put, and collar opportunities.

> Options Chain UI

MooMoo-style chain with expiration accordion, real-time Greeks, visual strike selection.

> Quote Fallback

Primary Moomoo API with automatic Yahoo Finance fallback. Never miss a quote.

> E2E Testing

Full Playwright test suite with GitHub Actions CI for regression and release gates.

moneyeraSingapore Property Analytics Platform
Live at moneyera.app
Live Site GitHub

"Bloomberg Terminal meets Google Earth" for Singapore property buyers. Map-first analytics with 4 financial lenses, AI-powered natural language search (Claude API), time-machine across 5 years of market data, draw-to-compare with radar charts, and commute isochrone overlays. Full-stack SaaS with Stripe subscription tiers.

React 18 TypeScript Python Hono Mapbox GL D3.js Recharts Cloudflare Pages D1 (Edge SQLite) R2 Claude API Stripe
MoneyEra — moneyera.app
Homepage PSF Heatmap Property Deep Dive Activity Lens
Homepage
Edge-First Architecture
DATA SOURCES URA API transactions data.gov.sg price indexes OneMap geocode / route PYTHON PIPELINE ETL Pipeline fetch → geocode → deduplicate → enrich amenities → compute aggregates → AI summaries CLOUDFLARE EDGE D1 Edge SQLite R2 raw / enriched Pages Fn Hono API (15+) Worker 6h reconcile FRONTEND (React 18 + Mapbox) 4 Lenses Time Machine Draw Compare AI Search Budget Map Integrations Claude API (summaries) Stripe (payments) Supabase (auth) Resend (alerts) Mapbox (tiles) Street View (images)

> 4 Financial Lenses

PSF heatmap, YoY appreciation, rental yield, and market momentum — all overlaid on the same map with instant switching.

> AI Natural Language Search

Claude API converts "freehold near CBD under 5M" into structured filters. Returns explanation of interpreted criteria.

> Time Machine

Temporal slider across 5 years of quarterly snapshots. Watch price waves ripple across the map.

> Draw to Compare

Freeform circle tool selects up to 8 properties. D3 radar + bubble charts for multi-metric comparison.

> Multi-Source ETL

Python pipeline ingests from URA, data.gov.sg, OneMap. Fuzzy deduplication, amenity enrichment, Claude AI summaries for 1000+ properties.

> Edge-First Stack

Entire backend on Cloudflare (D1 + R2 + Workers + Pages Functions). Atomic R2 manifest protocol for data consistency.

golf simulatorPhysics-Based Green Reading
Live on Vercel
Live Site

Interactive green reading simulator using Runge-Kutta 4th-order integration with USGA Stimpmeter calibration. Three modes: Learn, Practice, Quiz.

React 18 TypeScript Vite Zustand Vercel RK4 Physics
golf.suree.org
Golf Training Track Golf Practice Console
Training Track

> RK4 Physics

Runge-Kutta 4th-order integration for accurate trajectories. USGA Stimpmeter calibrated.

> Three Modes

Learn (guided tutorials), Practice (free exploration), Quiz (skill testing under pressure).

> Clock Face Method

Visual teaching tool showing how slope direction maps to break prediction.

> Real-Time Animation

Smooth ball physics with trail visualization. Watch your putt break in real time.