I'm a Pi-shaped engineer: deep technical expertise in backend systems and AI tooling (Go, TypeScript), active exploration of AI agents and the Model Context Protocol ecosystem, and product sense from shipping developer tools and iterating based on real-world usage.

jira-mcp — Go-based MCP server that makes Jira AI-native. Born from watching teams waste time on repetitive ticket operations. Now AI assistants can handle issue management, sprint planning, and workflow transitions directly. 76 stars, real teams using it in production. (Go)

obsidian-open-gate — Obsidian plugin that solves the isolation problem of local knowledge bases. Lets you seamlessly integrate web content into your notes without breaking your flow. 215+ stars from knowledge workers who needed this exact bridge between their notes and the web. (TypeScript)

InstantCode — Browser extension that eliminates context-switching when debugging frontend code. Click any element, ask AI about it—no copy-paste, no tab-switching. Integrated with GitHub Actions for continuous improvements from user feedback. 50 stars. (TypeScript)

mimaid — Minimal Mermaid diagram editor because existing tools had messy, overcomplicated interfaces. Built with real-time collaboration using Liveblocks and Yjs. Solving the DX problem: clean interface, fast editing, no clutter. Live at mimaid.aiocean.dev. (TypeScript)

copilot-sdk — TypeScript SDK for direct GitHub Copilot API access. For developers who want programmatic integration with Copilot in their own workflows without the editor middleman. (TypeScript)

autosocial-trends — Daily automated trend research pipeline. Runs in production, delivering fresh trend data without manual overhead. Built to solve the content creator's problem: staying current without burning time on research.

socketrpc-gen — Type-safe RPC over Socket.IO. Solves the DX gap where real-time communication breaks type safety, reducing runtime errors for teams using Socket.IO. (TypeScript)

AI-assisted development in production: Actively using Claude to ship code faster. mimaid uses GitHub Actions to automatically invoke Claude when builds fail—Claude diagnoses TypeScript errors, submits PRs, and fixes issues without manual intervention. Not just talking about AI coding assistants—using them daily to maintain and improve production apps. Testing what works: when to let AI handle routine fixes vs. when human judgment matters.

Production-grade automation: autosocial-trends delivers daily trend research through fully automated pipelines. Zero manual intervention, consistent execution across multiple data sources. Proving that well-designed automation beats manual processes for repetitive workflows.

Expanding AI integration surfaces: Building copilot-sdk to unlock programmatic GitHub Copilot access. Developers shouldn't be limited to IDE extensions—API-first access enables custom workflows, integrations, and automation that editors can't provide.

MCP adoption and iteration: jira-mcp (76 stars, 20 forks) demonstrates MCP's value for enterprise tool integration. Real teams using it in production validates the approach. Continuing to explore where MCP adds the most value in connecting AI to existing workflows.

• Backend & systems: Go for MCP servers, CLIs, and backend services. TypeScript for developer tooling and browser extensions
• AI tooling: MCP protocol implementation, agent orchestration, AI-assisted development workflows, building bridges between AI capabilities and existing developer workflows
• Real-time systems: Collaborative editing with Liveblocks, Yjs, and WebSocket-based architectures
• Type safety: Prioritizing type-safe APIs across Go and TypeScript—type-safety as a design requirement, not an afterthought
• Continuous delivery: GitHub Actions workflows that ship improvements based on real usage and user feedback. Testing AI-driven CI/CD patterns.

I build by identifying real workflow friction and shipping solutions:

• jira-mcp: seeing teams waste hours on repetitive ticket operations, gave AI assistants native Jira access
• obsidian-open-gate: watching knowledge workers manually context-switch between apps
• mimaid: frustrated with bloated diagram editors when all I needed was clean Mermaid editing with real-time collaboration
• InstantCode: debugging frontend code shouldn't require constant tab-switching
• autosocial-trends: manual trend research was blocking content creation, so I automated it
• copilot-sdk: developers wanted programmatic Copilot access without being tied to IDE extensions

Stars validate the problem exists. Daily commits and iteration prove the solution works. Production usage proves it's reliable.

• Ship early, iterate based on usage: InstantCode integrated GitHub Actions after seeing user patterns. autosocial-trends runs daily because that's what real users needed. mimaid deploys continuously to Cloudflare Pages with automated fixes.
• Technical depth meets product sense: Build MCP servers in Go for performance, but design APIs thinking about developer ergonomics. Type safety matters because it reduces user pain, not just because it's technically correct.
• AI exploration with engineering discipline: Testing MCP integrations, AI-assisted development workflows, and agent orchestration—but always with production reliability in mind. Not just talking about AI, actively using it to ship better code faster.

• Website: 12bit.vn
• Location: Việt Nam
• Company: Sen Việt
• Hireable: Yes — open to interesting projects in developer tooling, AI infrastructure, or workflow automation

I care about building tools that make developers faster and workflows smoother. If you're working on developer tooling, AI agents, MCP integrations, or solving real product problems in this space, let's connect.

Last updated: 2025-12-31