Talks

PyCon China 2026 welcomes speakers from various Python technology fields. Whether it's AI, agents, large models, embodied intelligence, physical AI, data science, web development, language features, engineering practices, cloud-native, automation, operations and maintenance testing, security, toolchains, quantitative trading, or more Python-based innovative practices and open source explorations, we look forward to you bringing real experiences, unique insights, and interesting stories to share with the community.

Please visit the link or scan the QR code below to submit your proposal: https://jsj.top/f/Nr10g4

Talks Lightning

Shanghai

2026-09-05

Integrated Building, Gubei Campus, Shanghai University of International Business and Economics

Production-Grade Python Agent Architecture: Topology Orchestration, State Persistence, and Secure Runtime Keynote

A rigorous software-engineering look at production-grade Python agent systems: event-driven multi-agent topology orchestration, distributed state and long-term memory persistence, and secure sandboxed high-concurrency runtimes.

Speaker

Yanjun Zhou - Solutions Architect, Google Cloud

Elasticsearch Semantic Search and AI Agent Development Keynote

A detailed introduction to semantic search in Elasticsearch and its latest developments. Concrete examples show how to use vector search (both sparse and dense vectors) for semantic search, how to easily create agents with Elastic AI Agent Builder, and how to integrate agents using the MCP and A2A protocols.

Speaker

When Organizations Have Memory: From Memory Engineering to AI-Native Organizations Keynote

From the five real problems of memory engineering, through harness optimization in the open-source terminal project con, to layered context continuity for multi-human multi-agent collaboration and building an organization that never forgets.

Speaker

Siwei Gu - co-founder & CEO, Nowledge Labs

Product Engineering in the Agent Era Keynote

Following a real product built from zero to one, this talk shows how agents participate in requirements, design, development, testing and deployment, and summarizes an agent-oriented product engineering methodology.

Speaker

Quan Yang - Founder of Tongsheng AI

Monitoring QEMU Guest State Using Python Keynote

This session introduces several mechanisms QEMU offers for communicating with the outside world, analyses the scenarios each one suits, and uses concrete Python script examples to demonstrate how to observe the state of a QEMU guest with Python.

Speaker

Yuchen Zheng - Practitioners in semiconductor-related industries

The Python Runtime as an Agent Interaction Interface Keynote

In this talk we will look at the kinds of Python capabilities agents need and their characteristics, at both traditional and emerging Python runtime implementations, at how these can be combined with existing agent frameworks, and at a comparison of how well the different combinations work.

Speaker

mew - Algorithm Engineer and AI Architect

The context management of code agent Keynote

This sharing will systematically guide everyone into the world of large model AI calling. It is suitable for friends who are interested in AI application development and want to understand the internal mechanisms of intelligent agents.

Speaker

Zhixiang Xu - A middle-aged fat man who can dive.

Contributing to CPython through Free-Threading Keynote

Drawing on my hands-on experience in the CPython Free-Threading community, this talk shows new contributors how to quickly set up a development environment, approach reading the codebase, find issues that suit them, and land a first high-quality pull request.

Speaker

Edward Xu - Senior Software Engineer, Citrix

A Python-Runtime-Based initramfs and Single-File Linux Implementation Keynote

An in-depth look at the design and implementation of the open source project mkinitramfs.

Speaker

Xinyi Shen - A Linux/Python user and student at Shanghai Jincai High School

AI-Guided Trading: Finding a Middle Way in Quant Keynote

A trading methodology for the AI era built on three pillars — factor mining, domain-partitioned management, and strategy fusion — combining quantitative signals with human judgment into a human-machine decision loop.

Speaker

Shihuan Zhang - Quantitative practitioner

How to Design the Institution & Platform for Agents Keynote

What makes an agent “one” individual, and why do we need many different agents? How should humans and agents collaborate? Drawing on the implementation at Dify.ai, this talk explores how to design institutions and platforms for agents.

Speaker

Yanli.one - Dify.ai

How to Implement Your Own Python DSL Keynote

Using a card-game DSL written from scratch as the running example, this talk walks through the whole picture: frontend code capture, IR design and lowering, optimization passes, and backend generation.

Speaker

Yan Lu - A slacker who chats during work hours and only works after hours.

Let Concurrency Be Free Keynote

A systematic exploration of concurrent runtimes, data-sharing boundaries, and the evolution of structured concurrency paradigms.

Speaker

Hongfu Wang - Backend Engineer

Skein: An Embedded Knowledge Database for the AI Era Keynote

Starting from why Skein was created, this talk shares thoughts on what databases should look like in the AI era, and the new needs, challenges and possibilities that appear when databases start serving “knowledge” rather than just “data”.

Speaker

Weizhen Wang - Co-founder & CTO, Nowledge Labs

Odoo: The WordPress of Enterprise IT Keynote

Through a comparison with WordPress, this talk introduces the 20-year journey of the Odoo open source project as Python’s killer application in enterprise IT, covering its evolution, open source logic, technical characteristics, ecosystem building and future prospects.

Speaker

Jianfeng Wang - Shanghai Open Source Information Technology Association Supervisor, Odoo China Community Initiator

Python Is More Than Writing Code: Reshaping Kindergarten Teachers’ Workflows with Python + AI Lightning

Starting from kindergarten teachers’ real work scenarios, this talk re-decomposes the teacher workflow with Python + AI and explores a human–AI collaborative workflow in practice.

Speaker

Wei Yue - Kindergarten technical support teacher

Who Decides the Next Step? Inside the AI Agent Loop, Tool Calling and Multi-Agent Collaboration Keynote

Using the open source Strands Agents SDK as a practical framework, this talk breaks down how an AI Agent actually runs — with Python code and live demos — so the audience leaves with a clear mental model of the agent execution mechanism.

Speaker

Yubin Zheng - Senior Developer Advocate at AWS

Stop Writing Interpreters in YAML: Rules as an Embedded DSL Keynote

Every system eventually grows a pile of decision logic that always breaks at runtime, at 3am. This talk offers another route: a type-safe, embedded DSL that lives inside Python.

Speaker

Nagato Yuzuru - TerraQuanta Backend and Infrastructure Engineer

Breaking Tool Lock-in: How the Agent Client Protocol (ACP) Connects Any IDE to Any Agent Keynote

Grounded in the Python ecosystem, this talk shows how ACP lets developers write code once and reuse their AI capabilities across mainstream editors like Zed, VS Code, and JetBrains.

Speaker

Jun Han - Senior Software Engineer, Microsoft Developer Division

How Many Steps Does It Take to Add LSP Support to a Python DSL Keynote

Starting from TileLang, we'll see how many steps it really takes to add LSP support to an operator DSL that looks like Python but doesn't write like it.

Speaker

Wenyang Wang - AI Inference Engineer, GitHub wanderer

From Compute to Intelligent Agents: MetaX's Open-Source Learning Projects Keynote

Centered on AI for Good, this talk traces MetaX's two open-source learning projects to show a complete path from underlying compute to intelligent agent applications.

Speaker

Zeyao Han - Head of Application Ecosystem at MetaX

Multi-Agent Harness for Long-Horizon Software Delivery Keynote

How Goals, Task Maps, dynamic Workflows, cross-provider Agents, controlled swarms, and independent verification turn scattered Agent sessions into an observable, recoverable, governable delivery system.

Speaker

Gongmin Liu - Agent Architect

Designing Plugin Systems for Python Packages Keynote

Drawing on real practice from PDM, this talk works through four mainstream plugin patterns to help developers choose the right architecture for their own libraries.

Speaker

Ming Xi - "Lying-flat" engineer

Making AI Capability Buildable, Reusable, and Collaborative: Developer Practice with SenseTime Raccoon's Skill and Expert-Team System Keynote

A new paradigm for large-model applications: moving from "tool calling" to "capability building," so expertise becomes reusable, composable, iterable AI capability units.

Speaker

Wangqun Pan - Senior Product Operations Manager at SenseTime, Head of Developer Ecosystem for the Raccoon family

Next-Gen Hadoop Storage: BlueField-3 and CSAL for Performance and Resilience Keynote

CSAL: open-source user-mode FTL, cache, and I/O trace component accelerating Big Data and AI storage within SPDK, commercially powering Alibaba Cloud.

Speaker

Wayne Gao - Principal Engineer and Storage Solution Architect at Solidigm

When Your AI Agent Dies at 3 A.M. - Reliability Engineering for Agents Keynote

Drawing on real incident records from the open-source AI Agent AnkaLoop, five production failure patterns and three fixes borrowed from databases.

Speaker

Jintao Zhang - CNCF Ambassador, Microsoft MVP

It's Time to Learn Rust — Starting with Making pip install Faster Lightning

This lightning talk is based on my real experience of teaching myself Rust as a Python developer. In 10 minutes I will cover three things: why it is worth learning, hands-on practice, and a one-month learning roadmap. No prior Rust knowledge is required.

Speaker

Zeyang Lin - Independent developer

Writing Custom Attention Kernels in Pure Python with FlexAttention Keynote

Writing a new attention variant used to mean writing a new CUDA kernel. Sliding window, ALiBi, document masking, PrefixLM. If your idea didn't match a kernel someone had already written, you were stuck with slow, memory-hungry PyTorch. FlexAttention removes that constraint.

Speaker

Himanshu Sangshetti - AI Engineer at Mem0

Profiling in CPython 3.15: Low-Overhead Tachyon Profiler for Real Python Systems Keynote

CPython 3.15 brings a clearer profiling story with Tachyon Profiler — a serious sampling profiler designed for finding real bottlenecks with low overhead.

Speaker

Petr Andreev - Russian Professor · CPython Contribution Team Lead · PSF Board of Directors Nominee

Let's Talk About the Boundaries of Data Validation — A Pydantic Rant Keynote

How can we perform data validation appropriately to strike the right balance between data consistency and service performance?

Speaker

Manjusaka Li - Freelancer, Open Source Enthusiast

Free-Threading Migration Guide Lightning

This talk discusses the issues involved in migrating existing applications to the free-threaded build: the code adaptations and thread safety work required, the expected performance gains and potential overhead, compatibility and upgrades of third-party libraries, and changes to the Python C API.

Speaker

Xiaowei Lu - Senior Software Engineer, Compiler Team, Alibaba Cloud

Xenix: A PySide6 Desktop AI Agent for Data Mining Lightning

A Windows desktop AI agent for data analysis and mining built with PySide6 for non-technical users at SMEs, enabling data processing, visualization, and machine-learning-based analysis through natural language.

Speaker

Yihong Yuan - Student at GDUFS

Let's Talk About Hash Functions in Python: Lessons from vLLM's Chained Hashing Evolution Lightning

Starting from the chained hashing design of vLLM Prefix Cache, this talk walks through the evolution from Python's built-in hash() to SHA-256 and xxHash, covering the trade-offs among performance, stability, and security.

Speaker

LuminolT - Graduate student at SJTU, member of TAI

From Selectors to Semantics: Vision-Enhanced Web UI Testing in Python Lightning

Building a vision-enhanced Web UI testing framework in Python with Playwright + pytest: locating elements and planning actions from screenshots and natural language, with a full AI-driven flow and MCP/Skill packaging.

Speaker

Zhiwei Chen - AI Engineer

From vector to cuVS Lightning

A decade of vector retrieval: from HNSW and Faiss to vector databases and RAG, to Faiss integrating NVIDIA cuVS — and the road toward lakehouse-native, unified retrieval engines.

Speaker

Hong Yi - An open source enthusiast

A Gentle Introduction to the GIL and Free-Threading Lightning

An accessible introduction to Python’s GIL and the Free-Threading build: explore the free-threaded world, understand both sides of the GIL, and decide whether you need Free-Threading.

Speaker

Muyao Yu - Student at Zhengzhou No.2 Senior High School

Your Coding Agent's history should become infrastructure Lightning

How to build a trustworthy session retrieval layer for agents and humans

Speaker

Yuu - Dify intern

Testing Open-Source Virtualization Infrastructure: A Practical Framework Lightning

Drawing on the ZSvirt Woodpecker testing project, this talk shares how to build an automated testing system for infrastructure that balances quality, efficiency, and community collaboration.

Speaker

Chao Zhou - Co-founder of ZSvirt, Product Planning at ZStack

Shenzhen

2026-09-13

10F Maker Center, Bank of Beijing Tower, MixC World, Nanshan District, Shenzhen

PyTorch Ecosystem in LLM Era Keynote

Centered around the PyTorch ecosystem in the large-model era and starting with the fundamentals of PyTorch training, this presentation sequentially introduces how Helion accelerates low-level operators, how DeepSpeed powers distributed training, how Safetensors stores and distributes model weights, how vLLM boosts the throughput of large-model inference, and how Ray orchestrates distributed tasks. It sorts out the full-stack open-source infrastructure maintained by the PyTorch Foundation, spanning from model training to production systems.

Speaker

Zhiqing Xiao - PyTorch Ambassador, PyTorch Foundation

Vericode: Formal Verification in the AI Era Keynote

Formal verification is becoming the ultimate safeguard for software quality — using mathematical proof rather than testing to guarantee that a program's behavior fully matches its specification. This matters especially in the AI era: when code is generated at scale by models, only formal verification can rule out hidden defects at the logical level. Lean4 is a formal language that combines strong expressiveness with efficient automation; built on dependent type theory, it can express programs, proofs, and mathematical definitions within a single logical framework. With AI-driven automation strategies — such as proof search, invariant synthesis, and automatic step completion — Lean4 significantly lowers the barrier to formal verification, giving critical systems machine-checkable correctness guarantees before release.

Speaker

Huang Yi - Freelancer

Xenix: A PySide6 Desktop AI Agent for Data Mining Lightning

Speaker

Yihong Yuan - Student at GDUFS

The Way of Agentic Lightning

Agents have gone through vibe coding, harness, loop, and graph cycles. The Way of Agentic decodes what lies beneath: goals, context, constraints, planning, verification and experience write-back — upgrading AI into a system that keeps compounding capability.

Speaker

Chao Peng - Chief Expert, ClawTime

Practical Local LLM Deployment Lightning

Real-world deployment experience on high-performance chips like the Apple M5 Max, covering hardware selection, quantization trade-offs, and tuning benchmarks.

Speaker

Shaofei Dai - Hengcheng IP