Boyuan LiBryan Lee

Full Stack & AI Platform Engineer building agent surfaces, gateways, automation, and data systems.

Full Stack & AI Platform Engineer github.com/liby
notes.looping.me
linkedin.com/in/lby
hi@looping.me
6+ yrs
Production software
1,194
Work MRs · 987 merged
5.8k★
Stars across open source
349
OSS PRs · 265 merged

Summary

Full Stack & AI Platform Engineer at RightCapital since 2019. I build the enterprise agent platform, AI gateway, automations, and the data paths behind them, from React streaming UIs to Dagster and Snowflake jobs. My default test for AI work is operational: routed, evaluated, metered, and loud when it fails.

Experience RightCapital · since 2019

2019Product UI

Built React and TypeScript product surfaces, then led migrations that retired years of tooling debt.

2025Platform systems

Moved across the API boundary into backend services, data pipelines, and AI infrastructure.

NowAgents

Own the surfaces and plumbing that let teams reach tools, knowledge, and models through one agent system.

Enterprise AI Agent Platform · internal product Platform architect
Role
The company's internal AI workspace: one assistant reaching enterprise tools and internal knowledge, adopted across the US and China offices.
Actions
Architected multi-model routing across Claude, GPT, and Gemini; OAuth integrations for GitLab, Jira, Notion, M365, and Salesforce; MCP support; E2B code execution; and unified search.
Impact
Became the company-wide agent surface, the single place teams reach tools and knowledge in natural language.
AI Gateway · platform Owner
Role
A single proxy in front of every LLM provider, so each team ships through one metered, observable interface instead of its own keys.
Actions
Built unified provider routing with token management, usage tracking, and an admin dashboard.
Impact
Serves all departments company-wide; centralizes the cost, quota, and audit that were previously scattered per team.
Event-driven Automation · Inngest Lead
Role
The connective tissue between GitLab, Slack, Jira and Salesforce.
Actions
Automated 10+ event-driven workflows, including AI code review, cross-platform user sync, email management, and Sentry aggregation.
Impact
Built on durable retryable steps, so failures surface and retry instead of disappearing.
Recording Intelligence & Ask AI · RAG Builder
Role
Two AI features touching the sales floor and customer self-service.
Actions
Added AI speaker classification and call summaries to the Recording Center, and launched RAG-powered Ask AI on the Help Center behind a quality-evaluation framework.
Impact
Speaker-labeled summaries reduce full-call review time; Help Center answers ship behind eval gates instead of subjective demos.

Open Source & Indie Work

The pattern is practical: find the failing layer, learn the toolchain, send the patch upstream. I taught myself Zig and Swift to repair the terminal I use daily, and maintain a translation plugin a large community relies on. Fixes also land in tools I build on: the Vercel AI SDK, shadcn/ui, uv, Dagster, and Inngest. 349 pull requests across open source · 265 merged.

Ghostty Learned Zig and Swift from scratch to fix performance bugs
★ 56k
Bob OpenAI Translator AI OCR translation for the Bob ecosystem · maintainer
★ 5.6k
dotfiles A dev environment and shared agent harness
★ 109
advanced-grammar A structured English grammar reference
★ 35

Engineering Judgment

2024Learn the stack the bug lives in

Ghostty performance work required Zig and Swift, so I learned both and merged 33 PRs into the terminal I use daily.

ongoingMake failure visible

A guard should fail at the bad state, not reroute silently. My reviews ask first: can this failure disappear?

2025Treat AI as infrastructure

I run a dual-stack review pipeline with an auto-fix loop and ship AI features behind evals. The useful work is the delivery mechanism.

Core Skills

Languages
TypeScript, JavaScript, Python, SQL, Rust, Zig
Frontend
React, Next.js, Tailwind CSS, Vite
AI & Agents
Vercel AI SDK, Agent Skills, MCP, RAG, multi-model routing, E2B sandboxes, eval frameworks
Backend & Data
Node · Bun, PostgreSQL · Drizzle, Inngest; Dagster, dlt, dbt, Snowflake, Azure AI Search, reverse-ETL
Practice
Root-cause debugging, multi-agent review pipelines, GitLab / GitHub CI, Sentry, Vercel, Cloudflare, Railway

Education

Southwest Petroleum University · Information Management & Information Systems
Northeast Forestry University · Art & Design