AI infrastructure engineer. I build the plumbing that makes AI agents actually work in production β agent frameworks, vector & knowledge-graph memory, MCP tool servers, and local-LLM integration. Mostly Go and Rust.
π Chengdu, China Β· π Remote Β· π¬ δΈζ / English
π’ Available for full-time remote & contract work β AI agents, RAG & knowledge graphs, MCP integrations, and LLM-powered automation. If you need an LLM wired into your real systems (not just a chatbot), let's talk β ll_faw@hotmail.com
- dataintelligence β β a governed semantic layer + MCP gateway that makes your data warehouse safe for AI agents. Agents ask for a metric by dimensions β never raw SQL β so wrong joins, wrong grains, and fan-out inflation are blocked structurally, not by prompting. Built on the libraries below: agent-go, cortexdb, eval-go, and semantic-go.
- semantic-go β the semantic-layer compiler underneath: declare metrics/dimensions/joins once in YAML, compile to fan-out/chasm-safe SQL.
- harness-rs β a Rust agent framework: ReAct loop, pluggable tools & skills, cross-session recall, a self-evolving learning loop, scheduler, sandbox, and an MCP client/server. Published on crates.io as the
harness-rs-*crates. - agent-go β an AI Agent SDK designed for Go developers (teams, tasks, memory, MCP, tool calling).
- tagit β an open-source, self-hosted Claude Tag: @mention an AI agent in your team chat, get auditable work back.
- oss-agent β a product-agnostic platform for AI ops & support agents over an OSS project: GraphRAG knowledge base, ReAct agent, and a deterministic red-line safety wall. The whole domain comes from one
domain.toml. - agentcli β an app-agnostic Go core for driving the Claude Code / Codex / Gemini CLIs: command building, stream-json parsing, usage accounting, PTY runner, hooks.
- eval-go β a native-Go LLM/RAG/agent evaluation framework: deterministic + LLM-as-judge metrics,
go testintegration, red-teaming, CI regression gates. The Go answer to DeepEval/RAGAS. - testtui β test TUIs, CLIs and agentic terminal apps via PTY + VT100 emulation: declarative YAML cases, an AI layer (harness-rs), and an MCP server.
- superai-desktop β a cross-platform Wails v2 desktop assistant and full showcase of agent-go: sandbox + browser + vision + autonomy + graph memory + skills, plus an SSE emotion protocol that drives external 2D/3D avatars (Live2D / VRM / Unity).
- aigui β a framework-agnostic TypeScript SDK that renders streaming LLM output as live UI: progressive markdown, cards, charts, math, and diagrams, with React / Vue / vanilla adapters.
- cortexdb β β a pure-Go, single-file AI memory & knowledge-graph library: vector + hybrid (BM25/FTS5) search, GraphRAG, and one-pass structured-data import. Zero external services, fully embedded.
- askdoc β ask questions over your own documents (RAG).
- mcp-swagger-server β turn any Swagger / OpenAPI spec into ready-to-use MCP tools.
- mcp-websearch-server β multi-engine web search with content extraction.
- mcp-sqlite-server Β· mcp-snapshot-server
- ollama-go β a Go client library for Ollama.
- lmstudio-go β a Go client for LM Studio: chat, embeddings, tool calling, model management.
- ollama-queue β a high-performance task queue for Ollama models.
- gosible Β· dispatch β infrastructure & multi-server automation in Go.
Go Β· Rust Β· TypeScript / React Β· LLMs (OpenAI-compatible, Anthropic, Gemini, Ollama, LM Studio) Β· RAG & vector search Β· knowledge graphs Β· MCP (Model Context Protocol) Β· SQLite
I'm best at taking an LLM from "demo" to "running unattended inside your systems": ingest your data, wire up the tools/APIs it needs (via MCP), and ship an agent that does the work β on a schedule, with guardrails.
π§ ll_faw@hotmail.com Β· π github.com/liliang-cn