Open Source Summit Korea 2026
2026-08-11..2026-08-12
Seoul, South Korea
[Open Source Summit Korea 2025, Open Source Summit Korea 2026]
as the organisers put it
Open Source Summit Korea unites the open source ecosystem, offering a platform for collaboration, knowledge-sharing, and exploring innovative solutions.
who it's for
open source maintainers, platform engineers, site reliability engineers, cloud engineers, security engineers, embedded developers, machine learning engineers
topics
OpenStack multi-cloud orchestration; Kubernetes LLM inference; WebAssembly on Kubernetes; kernel live patching; unikernel versus Linux; software bill of materials supply chain; observability for AI agents; AI-driven canary reasoning with Argo; GPU-aware load balancing for inference; Terraform drift detection with Falco; Zephyr RTOS lessons; V4L2 decoder driver implementation
| title | author | length |
|---|---|---|
| How x402 Brings Open Source Governance to Payments | Junhyeok Yoo | 35m |
x402 presents an open, chain-agnostic payments layer that uses HTTP 402 as a stateless challenge–response: a server returns payment terms and the client… | ||
| Panel Session: The Next Phase for Digital Money: From Pilots to Production in Korea and Beyond with Open Source | Julian Gordon, Alice Kim, Jisu Park, Joonyi Sung, Jake Kim | 27m |
Korea moves from PoC to production with Bank of Korea running a second phase since 2023 and major institutions (KSD, Koscom) trialing open‑source chains; focus… | ||
| Maximizing Heterogeneous Memory Bandwidth in Multi-Socket Systems | Rakie Kim, Honggyu Kim | 38m |
Presents Socket-aware Weighted Interleave implemented in the open-source HMSDK and upstream Linux patches to localize memory interleaving per physical CPU… | ||
| Self-Healing Rollouts: Automating Production Fixes With Agentic AI | Kevin Dubois | 33m |
A production demo shows an Argo Rollouts plugin that calls an external agent via the A2A agent protocol, compares canary vs stable (10% traffic, 10s warm‑up)… | ||
| Advanced Performance Profiling Using Perf Tools | Namhyung Kim | 37m |
Shows a set of perf-based tools that map PMU memory samples to DWARF to attribute accesses to specific structs, fields and cacheline offsets, with new perf… | ||
| Sponsored Session: From Clicks to Rankings: Building a Learning-to-Rank Pipeline with… | W. Kim | 31m |
An OpenSearch pipeline that gathers client-side click and query events using User Behavior Insights (UBI/UBIJS), joins them by query_id into query and event… | ||
| Bridge X86-64 Applications To ARM64 and RISC-V With Dynamic Binary Translation | J. Huang, C. Chiu | 20m |
Box64 runs unmodified x86-64 Linux binaries on ARM64 and RISC-V by preferring a wrapping-first architecture that routes calls into host libraries via ABI… | ||
| KNOD: In-Kernel Network Offload Device for GPU-Accelerated Packet Processing | T. Yoo, H. Lee | 28m |
KNOD is a Linux kernel framework that natively compiles XDP BPF into GPU ISA and dispatches packet processing without ROCm or userspace, using a Radeon RX 6600… | ||
| Sponsored Session: A New Security Standard for the AI Era: Zero-CVE and Real-Time Threat… | S. Lee | 32m |
The talk presents an automated, DevSecOps approach to achieving a Zero-CVE security posture for containerized workloads by hardening images with… | ||
| How I Tricked ArgoCD Into Sharding on a Single Cluster | Faeka Ansari | 32m |
ArgoCD sharding is cluster‑level only, so the fix is to make one cluster look like many by creating extra Kubernetes cluster secrets whose server URL differs… | ||
| From Black Box To Insight: Observability for AI Agents in Production | Mostafa Radwan, Brandon Kang | 30m |
A reference architecture shows traces, metrics and logs collected with OpenTelemetry and visualized in Prometheus and Grafana, deployed via Docker Compose… | ||
| The Butterfly Effect of a Broken Disk: Top-Down Ceph Troubleshooting To Upstream Contribution | S. Lee | 24m |
Small on-prem Ceph (10 nodes, 10 NVMe, 15 SSDs) experiences severe write latency traced with ceph osd perf, Grafana and dmesg -k to one node where an NVMe… | ||
| Building Event-Driven WebAssembly on Kubernetes: Runtime, Observability, and… | B. Kang, N. Hai | 31m |
Describes deploying WebAssembly workloads on Kubernetes using SpinKube and a Wasm runtime class (spin-v2), running ~150–200 randomized tests across six… | ||
| Sponsored Session: Trust by Design: Engineering Reliable Enterprise AI Workflows with Open Source | K. Kim, S. Woo | 37m |
An open-source reference architecture is built and demoed end-to-end: Docling and DocLang convert messy PDFs and tables into a machine-readable schema so… | ||
| LLMs Change — Where Should Knowledge Live? (Lessons From SBOM) | Koji Annoura | 28m |
LLMs produce useful answers but those answers change with model, prompt, and context, so knowledge needs to live as structured, persistent data outside the… | ||
| From Frankenstein To Kamaji: Lessons in Building a Single CAPI Cluster Across Multiple Providers | A. von den Driesch | 30m |
Built a hybrid workload cluster with control plane VMs on vSphere and bare‑metal workers on Proxmox, which triggers a deletion loop because the vSphere… | ||
| Implementing a Stateless V4L2 Decoder Driver: Architecture and Lessons Learned | SungHo Lee | 25m |
A stateless V4L2 decoder driver for Chips&Media VPU is implemented and upstreamed; GStreamer plugins and kernel driver live in mainline and support H.264… | ||
| Flexible Paging in Linux: Per-process Page Size | Dev Jain | 30m |
Design separates kernel and user page-size ABIs so processes can run with 64K pages implemented as 16 contiguous 4K pages while the kernel stays at 4K; mmap… | ||
| Beyond Round-Robin: GPU-Aware Load Balancing for LLM Inference in Kubernetes | Seokhwan Kong | 32m |
loxilb implements an eBPF-native AI gateway that uses XDP/TC and kernel sockmap for zero-copy L4 forwarding and adds L7 features (API-key, token-quota) plus… | ||
| What Happens When Your AI Agent Meets OPA | Jyoti Bisht, Animesh Pathak | 28m |
Shows concrete failures and fixes when an AI agent uses kubectl, cloud CLIs, Argo Workflows and MCPs. Examples: agent tries kubectl delete but instead renames… | ||
| From Region To Multi-AZ: Building Resilient Cloud Infrastructure With OpenStack, Kubernetes, and OVN | S. Han | 31m |
KT Cloud builds KCP—an OpenStack-on-Kubernetes platform (launched June 29) that uses OpenStack on Kubernetes with three role clusters (admin, management… | ||
| Keynote: Welcome | Jim Zemlin | 45m |
Conference opens by mapping where open source actually matters: infrastructure and models. Claims include $900B invested in 2026 and $1.4T capex in 2027 for… | ||
| SBOMs Aren't Enough. Secure Your Software Supply Chain End-To-End | Yongjae Chung, Justin Cappos | 37m |
SBOMs only inventory components and do not stop tampering; SLSA identifies nine distinct threat areas from source to distribution. The demo stitches gituff… | ||
| Computer Programming Is Dead; Long Live AI-First Programming | Stephen Chin, Cassandra Chin | 32m |
Argues programming as taught is obsolete and proposes an AI-first loop: try, learn, grow where learners prompt LLMs, inspect generated code, and extend real… | ||
| From Contribution To Culture: 14 Years of Building an OSPO That Outgrew Itself | Darae Ahn | 35m |
14-year OSPO journey begins with Tizen in 2012 and evolves into a culture change program that builds a core component program and an open source leader career… | ||
| The Era of Vibe Coding: Why High-Skill Engineers Are More Critical Than Ever | Yongjin Lee | 30m |
Builds a custom Linux OS called MaruxOS using Claude Code and shows concrete failures: an automated decision to replace libraries leads to a 70-file swap that… | ||
| Keynote: Linux in the Land of LLMs | Greg Kroah-Hartman | 18m |
LLMs detect bugs by pattern matching — Anthropic’s Mythos style scans reproduce old-vulnerability patterns against current code — and that drives CVE churn… | ||
| How AI Is Changing Open Source Communities: Lessons From openEuler | Jianmin Wang | 25m |
openEuler integrates AI across development: Intelligence BooM and earlier projects (AT, ops) automate performance tuning, package discovery and patch… | ||
| Stop Trusting a Black Box: The Economic Case for Open, Sovereign AI | Vincent Caldeira | 31m |
79% of organizations now treat AI sovereignty as strategic and 82% build customized AI solutions, motivating a move from rented APIs to reproducible stacks… | ||
| Upstream Kernel Hardening: Recent Progress and Challenges | Gustavo A. R. Silva | 35m |
Linux upstream hardening is shrinking long‑standing C hazards by combining source transformations, compiler changes and new helpers: enabling the… | ||
| Keynote: Open Source for AI and AI for Open Source | Priya Nagpurkar | 16m |
Argues that open models and tooling make enterprise AI controllable while AI reshapes open source by generating code and tests; cites a fivefold performance… | ||
| Keynote: Opening Address | Munyeong Lim | 05m |
Claims that national AI strength comes from ecosystem matters not just building models, and that open source can be a tool for world peace as well as… | ||
| Science in the Agentic Era: Structured Experimentation With Ado | Alessandro Pomponio | 26m |
ado is an open-source Python framework that constrains coding agents with schema-based validation (Pydantic), provenance tracking, and a structured datastore… | ||
| MCP Adoption and Why OSPO Skills Matter | Ana Jiménez Santamaría | 29m |
Model Context Protocol is an open protocol that standardizes connections between models and external systems using three primitives: tools, resources, prompts… | ||
| One Binary, Every Package Manager: Shipping a Rust CLI To PyPI, Npm, Homebrew, Winget, and Beyond | A. Kumar | 30m |
A Rust CLI named evnx for validating and secret‑scanning .env files is built, cross‑compiled into platform binaries and published to PyPI, npm, Homebrew… | ||
| GitAIOps: A 4-Layer Architecture for Predictable AI-Assisted Operations | Hoon Jo | 30m |
A 4-layer Git layout becomes persistent memory: Layer 1: Human plans in Git (36 files, 23,854 lines); Layer 2: Distilled AI context (6 files, 1,254 lines, 19:1… | ||
| Keynote: Open Models Need Open Memory | Nigel Brown | 03m |
Modern AI agents rely on a fast, durable, portable data layer that answers in milliseconds latency, survives restarts, and moves with deployments. The talk… | ||
| Personalisation and Specialisation of Search With OpenSearch Agentic Search | Aswath Srinivasan | 32m |
Demo of a secondhand luxury store shows the query "dress watch" rewritten by an OpenSearch flow agent with a query planner into OpenSearch DSL so results list… | ||
| Getting Started With Kernel Programming: The Linux Kernel Mentorship Program | Jori Koolstra | 26m |
Program guides newcomers to contribute 5–10 patches in 3–6 months, with weekly Q&A, Discord peer support, and co-mentors from core subsystems. Prerequisites… | ||
| Panel: Realizing Sovereign AI: Strategies for Korea’s Tech Sovereignty and AI Independence Via Op... | Yongkook "Alex" Kim, Priya Nagpurkar, Jae Lee, Alice Oh | 35m |
Panel lays out concrete steps for Korea to build sovereign AI: invest in hardware accelerators (GPUs, ASICs), networking and storage; run and fine-tune LLMs… | ||
| Open Source AI Agents on User-Owned Infra: K3s, MCP, & GPU Sharing in Practice | Miley Fu | 34m |
An open-source personal cloud OS runs on single-node K3s personal cloud, serves models locally (qu 3.6 27B tested vs. Claude 4.5 Opus), hosts private knowledge… | ||
| Keynote: Welcome Back | Jim Zemlin | 02m |
Nvidia presents its open frontier work including NeMoTron and wide contributions to PyTorch, claiming to be among the top open-weight model providers and… | ||
| Breaking the Sharing Paradox: A Working Foundation and a Call for… | Jinyong (Dave) Jeon | 22m |
OpenReagent presents a privacy-preserving signature format and pipeline that lets institutions share vulnerability signatures (not source code) so teams can… | ||
| From Closed To Collaborative: Perspectives and Lessons From Qualcomm’s Open Development Experience | Craig Northway | 31m |
Qualcomm moves a whole Linux product ecosystem from private downstream branches to a public, mainline-first model over 18 months, creating an Open Source… | ||
| Clouds on Clouds: OpenStack and Kubernetes With Cloud-Barista | Son Seokho | 40m |
Cloud-Barista provisions VMs on public clouds, boots bare-metal OpenStack instances on those VMs, registers that OpenStack back as a managed cloud, then… | ||
| Reboot Without Rebooting: Updating Linux Userspace at the Speed of a Service Restart | Nandakumar Raghavan, Prasanna Kumar T S M | 23m |
Soft‑reboot replaces the entire userspace by execv replacing PID 1 and pivoting to a new rootfs while the kernel and hardware stay untouched, implemented in… | ||
| Finding Vulnerabilities in IoT Embedded Devices Using Linux OS and Open Source Tools | Nkuba Kayembe Carlos | 37m |
A reproducible method for finding Z‑Wave flaws using only Linux, Python and a Yardstick USB dongle that fuzzes protocol fields rather than random bytes… | ||
| EZIO: Predictable, Fast, Scalable BitTorrent-Based Bare Metal Provisioning | Date (Yu-Chiang) Huang | 32m |
EZIO is a BitTorrent-backed bare-metal provisioner that transfers only used filesystem blocks and writes directly to raw disk on the fly, avoiding RAM… | ||
| From Static Rules To Reasoning Platforms: Scaling Intelligent Canary Delivery in 2026 | Daniel Oh | 35m |
Demonstrates an AI-driven canary pipeline that replaces static rules with a reasoning layer, wiring OpenAI/GPT-style LMs into Argo Rollouts and a Java Quarkus… | ||
| Connecting the Dots With Context Graphs | Stephen Chin | 23m |
Argues that persistent, queryable context beats one-off LLM calls and demonstrates a working pipeline: Neo4j + OpenAI embeddings + LangChain for a… | ||
| Exploring Unikernel: An Empirical Comparison With Linux | Taekyung Kang, Kyungha Kim | 28m |
Empirical comparison runs the same C app with musl on two Xen guests (ARM64) and measures build time, image size, and syscall/library latency with 5,000… | ||
| Who Watches the Watchers? Building Observability for the Platform Itself Across Multi-Cluster EKS | Faeka Ansari | 26m |
Platform observability for 11 EKS clusters across 6 AWS accounts uses labeled alerts (service, namespace, owner team, owner channel) plus trace IDs to link… | ||
| Kernel Live Patching: Mitigating CVEs With Zero Downtime | Shung-Hsi Yu | 18m |
Live patching applies function-level fixes without reboot by allocating new code, redirecting calls via ftrace trampolines, and switching tasks over only after… | ||
| Using AI To Bridge the Gap Between Safety Standards and Open Source Development | Kate Stewart | 47m |
AI extracts requirements and traceability from changing OS code, links them to tests and build SBOM metadata stored in SPDX 3.0+, and uses that graph to decide… | ||
| Keynote: Zephyr RTOS: 10 Years After Applying OSS Best Practices | Kate Stewart | 16m |
Zephyr RTOS claims measurable maturity: 61508 concept approval (Dec 2024) with 569 requirements and 873 linked tests tracked to source, active work toward… | ||
| Building and Orchestrating Production-ready Agentic AI Systems | Kevin Dubois, Daniel Oh | 31m |
Showcases a Java-based car-rental demo that actually runs: image and text inputs are analyzed by modular agents and routed to cleaning, maintenance, or… | ||
| Keynote: What Agentic AI Asks of Open Source Strategy: A New Layer for OSPOs | Ana Jiménez Santamaría | 15m |
Claim: agentic AI forces organizations to manage not just models but the surrounding open harness—instructions, tools, memory, execution loop, permissions… | ||
| Building Trust in the Age of AI: From Data Openness to Responsible Innovation | Dr. Honglak Lee | 17m |
LG AI Research reports building Korea’s sovereign foundation models (XL1 to XL1 4.5) and says trust hinges on dataset provenance, not just architecture or… | ||
| OpenChain’s Strategic Direction for 2027 in Global Practices | Meixia Wang | 31m |
OpenChain maintains ISO 5230 and ISO 18974 as its baseline, governed by 25 board members and adopted by 100+ companies, and delivers concrete tools: SPDX and… | ||
| From CVEs To Compliance: Automating Embedded Linux Kernel Security | Kyungsik Lee | 21m |
A CI-based vulnerability response pipeline automatically triages hundreds of kernel CVEs, applies clean upstream patches on a CI pass, and routes conflicting… | ||
| From Continuous Tracing To Automated Insights: AI-Driven Performance Analysis With Guider | Peace Lee, Jaeguk Lee | 37m |
Guider is a Python-based profiler that runs as a low-overhead daemon, collects kernel/user data via BPF, perf, ptrace, cgroups and Android hooks, and 200+… | ||
| Fusing AOSP and Linux! To Open a New Chapter for Linux Desktops | Yong Gong | 29m |
Claims that OpenFDE ports AOSP onto a standard Linux kernel using OXC containers so Android apps run natively on Linux with bare‑metal speed, addressing the… | ||
| Keynote: To Nemotron and Beyond: Building the Open Source AI Ecosystem | Chanran Kim | 13m |
NVIDIA claims open weights are driving an agentic open-source stack and lists concrete outputs: 650 open models and 250 open datasets on Hugging Face… | ||
| How To Make AI-Assisted OSS Contributions Review-Ready | Jaewoo Choi | 21m |
AI generates patches fast but maintainers still require reproducible environment, tests, and evidence before merging; a practical four-stage framework is… | ||
| Live Demo: Event-Driven Terraform Drift Detection With Falco | Keita Higaki | 33m |
Shows a real system that turns CloudTrail events into near‑real‑time Terraform drift alerts using Falco as the rules engine, a custom TF drift comparator… | ||
| Simplifying AI for Edge Compute with HAMi | Reza Jelveh | 28m |
HAMi provides a hardware-agnostic scheduling plane that carves GPU memory and compute so multiple agents run concurrently on small devices and time-shares when… | ||
| AI-Powered Open Source Risk Management: ISO Self-Certification Kit and 5-Level AI Coding Guide | H Jang | 33m |
Trusted OSS is an open-source (CC BY 4.0) kit that uses nine AI agents (Claude Code) to auto-generate 24 compliance artifacts—OSS policy, SBOMs, vulnerability… | ||
| When Semantic Search Breaks in your RAG systems: RAM Walls, Silent Failures, and the Architecture Decisions | Jeevan D C | 36m |
Team ships semantic search, hits 100M vectors, and RAM cost triples overnight because HNSW indexes live in memory: 1536‑dim vectors at 4 bytes ≈ 1 TB for 100M… | ||
| Building Reliable AI Agents: An Open Source Approach To Evaluation and Observability | Sho Tanaka | 30m |
Shows an open source stack (TruLens plus an AgentGPT‑style agent) that instruments agent workflows, captures execution traces and visualizes them in a… | ||
| Eval-Driven Development: Mastering Agentic Tracing and Expert-Aligned Judges With MLf... | V. Caldeira | 43m |
Claims a reproducible workflow that treats every agent run as an experiment, capturing telemetry with OpenTelemetry JAI semantic convention and storing… | ||
| nothing matches | ||