Ma Shouyue — Full-Stack AI Engineer
7 years of backend experience, from IM systems with 200k concurrent users to 100% AI-written Agent platforms. A full-stack engineer who has experienced product failure and independently completed full systems.
Tags: Ma Shouyue 27 years old Full-Stack Engineer AI Agent Freelance Remote Available
My Story
Section titled “My Story”Timeline
Section titled “Timeline”| Year | Event | Key Growth |
|---|---|---|
| 2019 | Graduated, developed smart parking system at a Beijing company | Hardware integration, backend CRUD, business understanding |
| 2020 | Changed jobs to a Beijing internet company, developed e-commerce system | K8S, Golang, microservices governance, observability, gray release |
| 2023 | Became an early OpenAI ChatGPT user, using GitHub Copilot | Pioneer practitioner of AI-assisted programming |
| 2023 | Released open-source project zero-im (IM system based on Go Zero framework) | Accumulated open-source community influence |
| 2024 | Company closed, gained clients through open-source project, started freelancing | From employee to independent developer |
| 2024 | Developed Telegram private server based on teamgram-server, supporting ~200k concurrent users | Large-scale distributed system practice, deep IM architecture understanding |
| 2025 | Failed at stock and crypto trading, lost all savings, but deeply practiced Vibe Coding | Product mindset awakening: technology is not the key, product is |
| 2026 | Returned to open-source, released RTC-Agent (100% written by Claude Code) | First product-mindset-driven Vibe Coding work |
From CRUD to AI Agent: A Programmer’s Awakening
Section titled “From CRUD to AI Agent: A Programmer’s Awakening”In 2019, I graduated and went to Beijing. My first job was developing a smart parking system. The project needed hardware integration, but most of the work was still CRUD. Back then, I thought backend programmers were just translating business logic into code.
In 2020, I changed jobs to an internet company developing e-commerce systems. That’s where I was guided to learn K8S, Golang, and open-source components I’d never used before: message queues, microservices governance, observability, monitoring alerts, gray release. But in reality, for backend programmers, it’s still mostly CRUD, understanding the business.
2023 was the watershed moment of my career.
That year, I became an early user of OpenAI ChatGPT. The first hottest OpenAI application was GitHub Copilot, and I was already using it. I remember it still required manually pressing the Tab key, but that shock of “AI helping you write code” made me realize: programming is about to change.
Also that year, the Go Zero framework started becoming popular. I built an open-source IM system based on it: zero-im. My colleague wrote the frontend + SDK, I handled the backend. This project gave me my first taste of open-source success — not because the technology was amazing, but because people were actually using it.
In 2024, life played a joke on me.
The company closed. But fortunately, my open-source project became popular. People asked me to customize IM systems, and I naturally became a freelancer.
That year, I built another brand-new IM — a Telegram private server, compatible with Telegram clients (Telegram clients are open-source). Someone had open-sourced teamgram-server in Golang, and I modified it to support most Telegram features. I also learned how Telegram does IM: it’s the update mechanism.
I helped clients maintain this system, supporting up to ~200k concurrent users online. A Go Zero microservices cluster, deployed on K8S.
That was my highlight moment: open-source project + commercial customization + large-scale system practice. I felt I had found my direction.
Then, in 2025, I collapsed.
I started trading stocks and crypto, using Vibe Coding to build trading systems along the way. The result was predictable — all failed, and I lost all my savings.
But that period gave me an important insight:
The key to Vibe Coding is not technology, but product.
If you haven’t figured out the product requirements yourself, haven’t thought through “who is this for? why would they use it?” then the product can’t be built.
I always start a Vibe Coding project thinking about the technology — no messy code, write code like Lei Jun writing poetry. This is the typical engineer’s disease. A product can’t be perfect from the start. When you find yourself stuck halfway and can’t proceed, you get lost. That’s how I used to be.
Today, I’m returning to the old path, continuing to open-source products.
RTC-Agent is my first product after this awakening. It’s a pure Vibe Coding product: 0% human-written, 100% Claude Code-written. All MVP features are now implemented.
This time, I’m no longer纠结 about whether the code is elegant or the architecture perfect. What I think about is: what problem does this product solve? who will use it? why use it?
If you’re also looking for a full-stack engineer who can independently deliver AI Agent systems, feel free to contact me.
Core Capabilities
Section titled “Core Capabilities”Protocol & System Design
Section titled “Protocol & System Design”Three-layer protocol architecture: authentication layer (OAuth2/HTTP) → operation layer (WebSocket RPC, 16 methods) → event layer (dual-channel Pub/Sub).
Distributed deployment solution: Nginx load balancing + stateful/stateless service separation.
AI Agent Engine
Section titled “AI Agent Engine”Built on Anthropic SDK + Eino, supporting Claude and OpenAI-compatible models.
Remote tool calling state machine + Redis Checkpoint crash recovery + infinite retry + exponential backoff, guaranteeing 100% tool delivery.
Frontend Engineering
Section titled “Frontend Engineering”Lit Web Components: 16+ sub-components, 9+ Controllers, @lit/context state distribution.
IndexedDB virtual file system + Babel AST script sandbox, dual-layer security isolation.
Security & Observability
Section titled “Security & Observability”OAuth2 dual Token mechanism, AST-level API blocking in script sandbox.
OpenTelemetry + Jaeger + Prometheus + Grafana + Loki full-chain tracing.
Tech Stack
Section titled “Tech Stack”Backend & AI
Section titled “Backend & AI”| Domain | Technology |
|---|---|
| Language | Go 1.27 |
| ORM | GORM |
| Vector Search | pgvector |
| AI Framework | Anthropic SDK, Eino |
| Script Sandbox | Babel AST transformation |
Frontend
Section titled “Frontend”| Domain | Technology |
|---|---|
| Component Library | Lit Web Components |
| Language | TypeScript |
| State Management | @lit/context |
| Frontend Storage | IndexedDB (Dexie.js) |
| Documentation Site | Astro 7, Starlight |
Infrastructure
Section titled “Infrastructure”| Domain | Technology |
|---|---|
| Database | PostgreSQL 17+ |
| Cache | Redis 7+ |
| Real-time Communication | Centrifuge WebSocket |
| Authentication | OAuth2 authorization code flow |
| Observability | OpenTelemetry, Jaeger, Prometheus, Grafana, Loki |
| Deployment | Docker Compose, Nginx |
Featured Project: RTC Agent
Section titled “Featured Project: RTC Agent”One sentence: Open-source Web AI assistant backend, through a standardized Remote Tool Calling protocol, lets websites integrate transparent, efficient, low-cost AI assistants with a few lines of code.
Core Innovation
Section titled “Core Innovation”AI infers on the server, but tools execute in the user’s browser. Data never leaves the user’s device.
| Metric | Value | Description |
|---|---|---|
| Integration Cost | 3 lines of code | Quick integration |
| Token Savings | 70%+ | Compared to screenshot+OCR solutions |
| Error Rate Reduction | 50%+ | Compared to server-side DOM scraping |
| Privacy Protection | 100% | Data stays in the browser |
Technical Highlights
Section titled “Technical Highlights”1. Remote Tool Calling Protocol
Section titled “1. Remote Tool Calling Protocol”- 6 built-in tools:
ls/read/write/grep/find/script - Complete state machine: Pending → Sent → Executing → Completed/Failed/Timeout/Rejected
- Redis Checkpoint crash recovery, infinite retry + exponential backoff
- 100% tool delivery guarantee
2. Virtual File System
Section titled “2. Virtual File System”- IndexedDB implementation, Dexie.js wrapper
- Directory structure:
/functions//scenarios//scripts//AGENT.md - Path traversal protection, large file paginated reading
- Automatic file type inference
3. Script Execution Engine
Section titled “3. Script Execution Engine”- Babel AST transformation + runtime permission confirmation, dual-layer security isolation
- Static API blocking: Storage / Network / DOM / eval all intercepted
- Pure computation standard library injection, built-in system tools (delay / uuid / now / random / time)
- Three modes: save / run / eval
4. Dual-Channel Real-time Communication
Section titled “4. Dual-Channel Real-time Communication”- Topic Channel: Persisted events + Offset tracking + offline recovery
- Live Channel: Redis Pub/Sub low-latency streaming output
- Epoch mechanism clears history, ensuring no message loss
5. Memory System
Section titled “5. Memory System”- Session Memory: 5 categories, up to 20 items, ~12K tokens
- User Memory: 4 categories, up to 1000 items, cross-session persistence
- Hybrid retrieval: vector cosine similarity + keyword full-text search, RRF fusion + importance weighting
6. Context Management
Section titled “6. Context Management”- Three-layer compression: Microcompact → Auto Compact → Session Memory Compact
- Configuration: 25K Token limit, 12K trigger threshold
- Auto Compact generates 9-part structured summary by LLM
7. Skill System
Section titled “7. Skill System”- Declarative (
agentConfig) + imperative (defineRegistry) dual registration methods - Function grouping, namespace paths
- Automatic documentation generation, Hook system (onStart / onSuccess / onError / onProgress)
- Scenario documentation: Markdown business workflow guides
Community Showcase: Mermaid Live Editor
Section titled “Community Showcase: Mermaid Live Editor”Integrated rtc-agent into Mermaid’s official online editor, less than 30 lines of glue code, zero modification to upstream source:
- 🗣️ Generate Mermaid code from natural language descriptions
- ✍️ Automatically write to editor and trigger real-time preview
- ✅ Proactively call syntax validation, self-repair on errors
- 🎨 No need for users to write a single line of code
Complete Documentation
Section titled “Complete Documentation”26 pages, bilingual Chinese-English, covering:
- Architecture overview, frontend, backend
- Core concepts: RTC, virtual file system, script engine, work modes
- Features: Session, messaging, Skill, commands, memory, context, real-time communication
- Integration guides: authentication, component API, function registration, scenario authoring
- Protocol reference: HTTP API, WebSocket RPC, events
- Deployment guides: source build, distributed deployment
What I Can Do for You
Section titled “What I Can Do for You”💬 IM System Development
Section titled “💬 IM System Development”This is my most experienced domain. From zero-im to Telegram private servers, I’ve built complete IM systems:
- Go Zero-based microservices architecture, supporting 200k concurrent online users
- Telegram protocol compatibility (update mechanism, message sync, group management)
- K8S deployment and operations, production-grade stability
🤖 AI Agent Systems
Section titled “🤖 AI Agent Systems”RTC-Agent is an AI Agent platform I built from scratch:
- Self-developed Remote Tool Calling protocol, letting AI execute operations in the browser
- Agent engine + memory system + context management + script sandbox
- Anthropic SDK / Eino integration, supporting Claude and OpenAI-compatible models
- Complete frontend component library (Lit Web Components) + backend services + documentation site
🔧 Go Microservices + Infrastructure
Section titled “🔧 Go Microservices + Infrastructure”Years of backend practical experience:
- Go + GORM + PostgreSQL + Redis tech stack
- WebSocket real-time communication (Centrifuge)
- OAuth2 authentication, microservices governance, observability (OpenTelemetry + Prometheus + Grafana)
- Docker / K8S deployment
🌐 Full-Stack Development
Section titled “🌐 Full-Stack Development”- TypeScript frontend + Lit Web Components component library
- IndexedDB local storage, Babel AST script sandbox
- Astro documentation sites, bilingual Chinese-English technical documentation
Why Choose Me
Section titled “Why Choose Me”- Independent Delivery Capability: RTC Agent from protocol design to UI components, from Agent engine to documentation site, all independently completed
- Deep Tech Stack: Not just API calling, but understanding underlying mechanisms (state machines, Checkpoint, AST transformation, dual-channel communication)
- Product Mindset: Focused on user experience and developer experience, not just task completion
- Privacy First: Privacy as a design constraint, not an add-on feature
- Complete Documentation: Delivering not just code, but complete documentation, examples, and integration guides
Contact
Section titled “Contact”- 📧 Email: [meishouyue@gmail.com]
- 🔗 GitHub: @PineappleBond
- 🌐 Website: RTC-Agent
- 💬 Phone: [+86 15666355528]
If you’re looking for a full-stack engineer who can independently deliver complete systems — from protocol design to frontend components, from Agent engine to K8S deployment — feel free to contact me.