Engineer & independent researcher

Pengcheng Xu.

Engineering at Speak.
Currently building AgentConnect ↗.

Selected projects
Based in Shanghai, ChinaA little about me Compilers · Runtimes · Agents

About

I work where systems engineering
meets independent research.

I work in engineering at Speak and am currently building AgentConnect, an open-source platform for teams and AI agents to work together. My background spans compilers, distributed runtimes, streaming, and developer tooling — with a focus on reliable infrastructure around increasingly capable models.

Engineering at Speak since July 2026. Previously Microsoft and Elastic.

Current explorations

  • LLM agent systems & orchestration Agentic workflows, tool-use, supervisor graphs
  • Inference-cost & latency optimization Making autonomous agents cheaper and faster to run
  • Compilers, IR/DAG & code generation For streaming and data systems
  • Runtime sandboxing & secure execution Multi-tenant isolation for untrusted workloads
  • Developer tooling & program intelligence Release intelligence, code navigation, build & delta analysis

Projects

Open-source tools and research spanning LLM-agent systems, developer tooling, and compilers.

Explore the projects

Selected work & research

AgentConnect ↗Current project · Open source

A platform for teams and multiple AI agents to collaborate in shared conversations and workflows. Connect agents to Slack, GitHub, and other tools, with shared context, permissions, and self-hosted execution.

Multi-agent systems · orchestration · developer infrastructure

Eventstreams V3 Topology Re-architecture Microsoft Fabric · 2023–2026

As acting Tech Lead, re-architected the core real-time streaming engine onto a unified backend compilation pipeline and a high-performance IR/DAG framework. Serves Fabric customers globally at ~5M events/day with ~180% YoY growth.

Compiler / IR / DAG · codegen · distributed streaming

Sentio.xyz Agent Orchestration Open-source contributor · 348+ commits

Designed and shipped a LangGraph supervisor-graph agent orchestration system; built an LLM processor code-generation agent with multi-chain support (EVM, Sui, Aptos, Move); owned the EVM tracer, debugger, and Solidity language server.

LLM agents · supervisor graphs · multi-chain codegen

NL → Streaming-Pipeline Agent Microsoft org hackathon · 2nd place · Oct 2024

An LLM agent that compiles natural-language business constraints directly into deployable Fabric Eventstreams pipelines.

LLM agents · program synthesis · streaming

Background

SpeakJul 2026 – Present

Engineering

Microsoft Apr 2020 – 2026 · Beijing → Shanghai

Technical Lead / Senior Software Engineer

Fabric Eventstreams (acting Tech Lead), Azure Stream Analytics core engine, and Azure Synapse Spark notebooks.

Elastic May 2018 – Apr 2020 · Beijing

Software Engineer

Codesearch team (joined via the Insight.io acquisition); built a high-scale Java language server on Eclipse JDT and LSP code-intelligence in Kibana.

YITU Technology Jul 2017 – Apr 2018 · Shanghai

Software Engineer Intern — Infrastructure

Infrastructure and tooling across a high-availability inference platform.

Upstream open-source microsoft/Trill tikv/tikv elastic/kibana

Education

University of Science and Technology of China

B.Sc. Computer Science & Technology

2014 – 2018 · USTC

Skills

Languages

C#PythonGo RustTypeScriptJava C++SQLScala

LLM & Agents

LLM agent designAgentic workflows Prompt/context engineering Inference-cost optimization

Systems

Distributed systemsCompiler construction LSPIR/DAG Runtime sandboxingHigh-throughput streaming

Infrastructure

Azure FabricStream Analytics SynapseOneLake ElasticsearchDocker LinuxCI/CD