Software Architect & Tech Lead · 10+ yrs building large-scale systems · now deep on LLM & agentic AI

I own technical direction end to end. Over 10+ years I've been the architecture owner and technical lead across multiple concurrent client products — scoping with stakeholders, making the architecture calls, and driving delivery on systems that serve millions of users at 99.9%+ uptime. Deep on microservices, real-time platforms, and cloud-native systems on GCP/AWS.

Architect  →  designs the system
Tech Lead  →  ships it with the team
AI builder →  writes the agents and tooling underneath

• 🔭 Currently: building LLM & agentic AI at Mobilefolk — production RAG, fine-tuning (LoRA/QLoRA), context engineering, eval/observability
• 🌱 Going deep on: agent orchestration, MCP, retrieval that actually beats grep
• 🛠️ Hands-on daily with Python, TypeScript, and agentic coding tools (Claude Code, Codex)
• 💬 Ask me about: system architecture, RAG pipelines, fine-tuning, or building MCP servers
• 📫 Reach me / portfolio: microvn.net · founder of OkLabel, an AI data-labeling platform

MCP-native code-context engine for AI coding agents. Pre-computes the call / import / override graph so agents see true blast radius instead of grepping the whole repo.

2.0× BM25 quality  19% fewer tokens  ~14ms queries

Beats every other MCP code-graph server head-to-head: GitNexus, CodeGraphContext, Codebase-memory-mcp, and Code-review-graph.

Spec-first agentic dev toolkit: ~12 commands enforcing spec → tests → build. Works with Claude Code, Codex, and other coding agents.

8 language stacks  shipped to 20+ engineers

Annotation layer for docs from any source — AI-generated, hand-written, exported. Publish a doc as a link, collect margin comments, and pull that feedback back into a coding agent over MCP, all on your own box.

MCP feedback loop  single docker compose

AI / LLM  _{RAG · fine-tuning (LoRA/QLoRA) · context engineering · OpenAI API · MCP}

Languages

Backend & frameworks

Data & infra