AI systems builder focused on LLM infrastructure, agent/eval platforms, and open-source contribution discipline
- 🧠 I build AI systems across LLM infrastructure, retrieval, agents, evals, and backend platforms
- ⚙️ Strong focus on production GenAI systems, AI infrastructure reliability, and developer tooling
- 🌱 Currently building a weekly open-source contribution track across NVIDIA NeMo, MCP/agent infrastructure, Microsoft agent tooling, vLLM, PyTorch, and LangChain/LangGraph
- 🤝 Contributing toward source-backed, test-backed, maintainer-friendly open-source AI infrastructure
- 💡 Developing Agentic Contribution Kit, a portable method for turning GitHub issues into responsible AI-agent-assisted PR packets
- 🟦 NVIDIA-NeMo/Automodel — Draft PR #2730 for issue #533: adds a DP-aware
StatefulDataLoaderwrapper so checkpoint resume can reshard dataloader progress when data-parallel world size changes, with focused CPU unit coverage for DP scale-up, scale-down, same-DP restore, Megatron batch samplers, and checkpointer rank-state loading. - 🟦 NVIDIA-NeMo/Automodel — Open PR #2627: implements safer consolidated SafeTensors shard writes for Databricks/Unity Catalog workflows, with regression coverage and documentation updates.
- 🟪 IBM/mcp-context-forge — Open PR #5329: standardizes root-path-aware admin and SSO redirects for issue #1588, routes security middleware through the shared root-path resolver, hardens mocked-settings fallback handling, and adds focused admin, SSO, security-header, and path utility tests.
- 🟦 microsoft/hve-core — Review-ready PR #2096: updates the Agent Systems Catalog after the
architecture-diagramscapability moved from a dedicated agent to a reusable skill, with Copilot review feedback addressed and maintainer review pending.
- 📌 Current targets: NeMo AutoModel checkpointing; IBM ContextForge MCP gateway routing; Microsoft HVE Core agent documentation
- 🔍 Focus: LLM infrastructure, SafeTensors consolidation, MCP gateway path handling, security middleware, agent workflow documentation
- ✅ Style: issue-first contribution, focused patch, regression tests, reviewer-feedback iteration, DCO sign-off
- 🧩 Agentic Contribution Kit — AI-infra contribution intelligence for LLM systems: repo knowledge spines, paper-to-repo hypotheses, eval gates, reward loops, and PR packets
- 🚧 Project slot 2 — placeholder for the next public AI lab project
- 🚧 Project slot 3 — placeholder for the next public AI lab project
- 🚧 Project slot 4 — placeholder for the next public AI lab project
- 💻 Languages: Python, SQL, JavaScript/TypeScript
- 🤖 AI systems: PyTorch, vLLM, Transformers, LangChain, LangGraph, MCP
- 🔎 Retrieval and evals: RAG pipelines, vector search, metadata filters, eval sets, tracing
- 🌐 Serving and platform: FastAPI, Docker, cloud deployment, observability, latency/cost tracking
- ✅ Quality: pytest, source-backed reviews, regression tests, PR validation, CI-aware development
- 🚀 I turn AI-agent development into reviewable engineering packets, not just generated patches
- 🔍 I like source reading, duplicate checks, validation, and reviewer-feedback loops as much as implementation
- 🧩 I combine GenAI product experience, backend/platform engineering, and ML systems learning
- 📈 I am building public proof gradually through small, useful, validated open-source contributions
- 💻 GitHub: github.com/huahuajhu
- 💼 LinkedIn: linkedin.com/in/h-hua-a9460042
- 🤝 Open to AI systems, GenAI platform, ML infrastructure, inference/eval tooling, and open-source collaboration