AI_dev Europe 2025
2025-08-28..2025-08-29
Amsterdam, Netherlands
| title | author | length |
|---|---|---|
| Welcome + Opening Remarks | Stephen Chin | 05m |
Conference frames open source as the default for AI with 63% of organizations using open source AI models and a survey by Misole Foundation, McKenzie and the… | ||
| 3 Pillars of Open Source AI: Training, Inference, Agents | Mark Collier | 42m |
Open-source AI splits into three concrete pillars: training (PyTorch holds ~80% market share, big projects scale from 20B to 120B parameters), inference… | ||
| The AI Cocktail | Malte Pietsch | 17m |
Company built two products (Haystack and deepset AI platform) and reports working with hundreds of teams and 400,000 monthly active developers. Key claim… | ||
| Zero to 50 ExaFLOPS in Under a Year: Lessons from the Trenches | Hagay Lupesko | 17m |
Cerebras scales an inference service to 50 exaflops in under a year by deploying wafer-scale engines with 900,000 cores and 125 petaFLOPS per chip across… | ||
| The Post-Modern Data Stack for AI: Open Source Foundations for a Multimodal La... | Chang She | 16m |
Lance format is a new open source lakehouse storage that stores images, audio, video, embeddings and tables in the same table on object storage, designed for… | ||
| Welcome Back | Stephen Chin, Hilary Carter | 08m |
Linux Foundation Research presents specific, evidence-backed claims: AI is a net job creator, with organizations 2.7 times more likely to expand headcount than… | ||
| OpenSearch and the Path to a Trillion-s... | Vamshi Vijay Nakkirtha, Navneet Verma | 26m |
OpenSearch presents concrete engineering changes to reach trillion-scale vectors: 3x search throughput, 3x better disk storage utilization, and 10x faster… | ||
| Frontier AI Models: A Review of Past Nine Months | Ozgun Erdogan | 25m |
Survey finds reasoning models like o3 and DeepSeek R1 move workflows from iterative chat to single‑shot problem solving by fine‑tuning on stepwise reasoning… | ||
| Vibing with Data -- Multi-Agent Data Modeling and Constructi... | Andreas Kollegger | 36m |
Describes a concrete multi-agent pipeline that takes files to a runnable knowledge graph by iterating: clarify intent, discover relevant CSV/markdown, propose… | ||
| Beyond Code Completion: The Shift of AI Programming from Pair to Peer | Nicky Pike | 24m |
Coder runs an internal agent called Blink inside cloud development environments and turns it loose on real work for 2.5 months, producing 179 successful issue… | ||
| Securing AI Pipelines: Real-World Attacks on Kubernetes-Based AI Infrastructure | Abhinav Sharma | 25m |
Deploys Stable Diffusion from Hugging Face onto AKS using Kubeflow pipelines built from Jupyter notebooks (kfp, diffusers, transformers, torch), stores models… | ||
| The Art of Production Quality Retrieval | Chang She, LanceDB | 29m |
Walks through a production retrieval pipeline: generate captions, mood keywords and image/text embeddings, parse queries with an semantic router LLM to route… | ||
| From Cold Start To Warp Speed: Triton Kernel Caching With OCI Container Images | Maryam Tahhan | 20m |
Proposes packaging Triton-generated LLVM kernel caches into signed OCI container images to prepopulate nodes and bypass JIT compilation, implemented as MCU… | ||
| Unlocking Networking AI Innovation: Introducing Essedum | Arpit Joshipura | 15m |
Essedum is an open-source Essedum framework that stitches MLOps, cloud connectors and accelerator APIs via REST into deployable network AI apps, donated to LF… | ||
| From Hours To Milliseconds: Scaling AI Inference 10x With Serverless on Kubernetes | Anmol Krishan Sachdeva, Paras Mamgain | 37m |
Claims a real deployment path that cuts latencies from seconds to under 100 ms and boosts throughput about 10x by running models serverless on Kubernetes… | ||
| How We Built Our First MCP Server: Lessons From the Trenches at Hud | May Walter | 27m |
Hud builds a production MCP server that connects live runtime signals to code by computing a function identity (no source stored) and matching it to the… | ||
| Building a Conversational Knowledge Base | Kerim Satirli, Tu Nguyen | 30m |
Built a conversational knowledge base that indexes robot docs and specs and returns focused answers using retrieval-augmented generation with embeddings and a… | ||
| From Planning To Production-Ready RAG | Ezequiel Lanza, Andreas Kollegger | 01h 00m |
Demonstrates a deployable edge RAG stack built from the Open Edge Platform: modular AI libraries and AGI suites (image/video search, robotics), model servers… | ||
| Sovereign AI: Building... | Anna Hermansen, Hilary Carter, Adrienn Lawson, Anni Lai | 27m |
Research shows sovereign AI is driven by localization, cultural values and security needs and is already a strategic priority: 79% view it as important and 93%… | ||
| Scalable LLM Inference on Kubernetes With NVIDIA NIMS, LangChain, Milvus and Flu... | Riccardo Freschi | 33m |
Blueprint for a retrieval‑augmented chatbot running on Amazon EKS that actually deploys embedder and LLM as NVIDIA NIMS, stores vectors in Milvus (uses… | ||
| Streamlining AI Pipelines With Elyra: From Development To Inference With KServe & VLLM | Ritesh Shah | 26m |
Elyra converts Jupyter notebooks into executable Kubeflow DAGs so a data scientist drops in Python cells and Elyra converts notebooks to Kubeflow pipelines… | ||
| Fast Inference, Furious Scaling: Leveraging VLLM With KServe | Rafael Vasquez | 23m |
vLLM claims to cut KV cache memory waste from 60–80% to about 4% by using vLLM paged attention, plus continuous batching and custom CUDA kernels to boost… | ||
| Unlocking Scalable Distributed Training With Arrow Data Cache on Kubernetes | Ricardo Aravena | 26m |
An Arrow-based cache sits between object storage and training, decoupling preprocessing and serving columnar batches with zero-copy Arrow Flight so… | ||
| AI-Powered Search in Modern E-Commerce | Stanko Kuveljic | 33m |
Claims a production-ready pipeline that combines precompute product embeddings with query-time encoder (≈50 ms) and a multi-stage flow that sends top-N vectors… | ||
| Docling: Get Your Documents Ready for Gen AI | Michele Dolfi, Peter Willem Jan Staar | 31m |
Open-source Docling converts PDFs, DOCX, PPTX, HTML, images and Markdown into a structured DocklingDocument unified format, extracts layout, tables, figures… | ||
| Beyond Prompts: Building Intelligent Applications With Genkit and the Model Context Protocol | Peter Friese | 37m |
Genkit is an open source AI integration framework that runs on Node/TypeScript, Go and Python and can deploy anywhere (even a Raspberry Pi), providing Genkit… | ||
| What's Coming Next After OSAID V.1 | Stefano Maffulli | 37m |
OSI releases OSAID v.1 in October 2024 and reports a yearlong coalition of 100+ contributors from 27 countries that defines open-source AI as granting freedoms… | ||
| Reverse Engineering Using LLMs | Vutukuri Sreenivas | 26m |
Claims that LLMs convert COBOL to Python/Java quickly using a pipeline of tokenization, semantic chunking, fine-tuning and CI/CD, with CodeLlama and Hugging… | ||
| Breaking RAG Systems: Exploiting Vulnerabilities & Hardening Your GenAI Applicatio... | Abhinav Sharma | 26m |
A working RAG prototype uses OpenAI embeddings, GPT‑3.5‑turbo, Chroma vector store, LangChain QA chains and a Streamlit UI and is successfully broken via… | ||
| Trallie: Shaping Unstructured Data Into Valuable Information | Vijayasri Iyer, Cristiano De Nobili | 27m |
Open-source framework Trallie converts file collections (PDF, HTML, TXT, JSON) into validated JSON using LLMs (GPT‑4o, Grock, Llama, Mistral, Hugging Face… | ||
| Turning Emissions Data Into Climate Actions With CityCatalyst | Mirco Rudolph | 28m |
CityCatalyst is an open-source platform that builds GHG inventories aligned with the GPC Protocol and outputs AI-ranked mitigation and adaptation options. The… | ||
| When Kubernetes Meets MLflow: Orchestrating MLOps Processes the Cloud Native Way | Diego Braga | 25m |
Demonstrates a concrete pipeline: MLflow configured and controlled by Kubernetes operators so the cluster becomes the reconciliation plane. Uses MLflow… | ||
| Observability Without Overs... | Joaquin Rodriguez, Amin Espinoza de los Monteros | 54m |
Hands-on workshop shows a CLI chatbot + backend running a local Olama 54-mini model, wires telemetry through an OpenTelemetry collector and OpenLIT SDK, and… | ||
| InfiniEdge AI: Bringing Open-Source Generative AI To the Edge | Ezequiel Lanza | 31m |
InfiniEdge AI is an LF Edge open‑source framework that already ships two releases and integrates five new subprojects to run generative models on tiny devices… | ||
| LLM-generated Code and Open Source License Compliance: How Big Is the Problem? | Oscar Enrique Goñi | 31m |
Using a 35× larger reference set than Xu et al. — osskb.org (≈27 TB) queried via scanoss.py — the study finds LLM outputs match existing open‑source code at… | ||
| Arctic Inference: Open-Source LLM Inference Optimizations at Snowflake | Aurick Qiao | 28m |
Arctic Inference is an open‑source vLLM plugin that implements shift parallelism, speculative decoding, suffix decoding and SwiftKV and wires them into vLLM as… | ||
| RAG at Scale: Logging, Traceability, and the Architecture for Control | Alison Cossette | 31m |
Live production RAG uses Neo4j to store 15,000 chunks (512-byte chunks with ~20–30 overlap), builds kNN edges, runs kNN + Louvain community detection, and… | ||
| AGENTS OF S.E.A.L.E.D: AI Agentic Cybersecurity Framework | Krishnendu Dasgupta | 37m |
Modular on‑prem agentic framework runs as ~12 Docker containers and orchestrates models across 6GB–24GB GPUs, using Gemma 270M intent model to pick services… | ||
| From Radar Echoes To Real-Time Recognition: Deep Learning on Another Planet -An Open Source Workflow | Viktor Somogyi | 31m |
Pipeline converts Magellan SAR mosaics into versioned GeoTIFF/GeoJSON layers that geologists can measure and query, with tile prep, speckle-aware augmentation… | ||
| AI Agents for Multiprovider Cloud Edge Continuum | Gentiana Canga, Paolo Fasano | 22m |
A deployed microservice orchestration that accepts natural‑language service requests, looks up apps in a vectorized repository, builds deployment intents and… | ||
| EVE: An Open Source Earth Science LLM for Researchers, Policymakers, and the Public | Àlex R. Atrio, Vijayasri Iyer, Marcello Politi | 34m |
EVE fine-tunes Mistral 3.2 24B on a curated EO corpus (about 5.3B open tokens plus 1.1B closed tokens), creates ~1B synthetic QA tokens grounded in retrieved… | ||
| Scaling Test-time Inference Compute & Rise of Reasoning Models | Jayita Bhattacharyya | 32m |
Claims that an open reasoning stack (Deepseek v3 → R1) turns a 671B base into a reasoning model using prompting with <think>/<answer> tags, two reward signals… | ||
| Emotion Estimation for Social Robotics | Samer Attrah | 21m |
System converts MediaPipe landmark output into 52 floating MediaPipe blendshapes, prunes them to 27 and feeds sequences into an LSTM 4-layer + FC classifier… | ||
| Vector Search Made Simple: Getting Started With OpenSearch for AI Applications | Dotan Horovits | 22m |
OpenSearch functions as an end to end vector database that stores dense embeddings, supports exact and approximate k-NN (HNSW, IVF), and scales to trillions of… | ||
| BAML - A New Programming Language for Building Agents | Vaibhav Gupta | 15m |
BAML is a new domain-specific language for agent code that treats prompts as typed functions, provides built-in test cases, visualizes raw model requests and… | ||
| Build for Speed: Lessons from Empowering GenAI Research at Canva | Verena Dittmer | 15m |
Canva runs GenAI research with over 200 engineers, supporting 100+ models and about 2k training runs/month, aiming for “Demo in a Day” prototypes via Gradio… | ||
| The Future for AI-First Enterprises is “Open” - How to Succeed and Not... | Manos Koukoumidis | 10m |
Enterprises that stay on closed-model APIs face overspending by more than 10x and limited customization; open models close the quality gap (from ~8% to ~1%+)… | ||
| The Next Compute Revolution: Are We Ready for Quantum? | Jennifer Prendki | 15m |
Claim: useful quantum machines are nearer than the “30 years away” refrain and already power niche wins in search and optimization, quantum chemistry and… | ||
| From POC to Profit: Deploying Customer-Facing AI Agents That Actually Work | Zhou Yu | 18m |
Claims that 90% of customer-facing agents stay stuck in POC and that production requires ~50× more work, plus a platform that creates synthetic users and… | ||
| The Imperative for Open Source AI - The Path Forward to Take Down the AI Oligarchs | Emmanouil (Manos) Koukoumidis | 22m |
Argues that open frontier AI is inevitable and details a working platform that makes it practical: an Apache 2 codebase with SDK and CLI, config‑driven… | ||
| Declarative Device Virtualization: Orchestrating GPUs & Hardware in Cloud Native Environments | Samrat Priyadarshi, Anmol Krishan Sachdeva | 34m |
Explains why MIG, MPS and time‑slicing trade isolation, latency and flexibility and shows a software alternative: Hammy (HAMI) CNCF sandbox that intercepts… | ||
| Agentic GraphRAG: AI’s Logical Edge | Stephen Chin | 24m |
Don't miss out! Join us at the next Open Source Summit in Seoul, South Korea (November 4-5). Join us at the premier vendor-neutral open source conference… | ||
| Building AI Workflows: From Local Experiments To Serving Users | Oleg Šelajev | 25m |
Claims a repeatable path from local LLM hacks to deployable apps using containerized, static configs and shows concrete tooling: Docker model runner for local… | ||
| Context Engineering: Practical Techniques for Improving Agent Quality Today | Vaibhav Gupta | 26m |
Claims that an LLM is best treated as LLM = function and tool calls are simply data extraction into unions, with agents implemented as agent = while loop +… | ||
| Defining Open Source AI: Can the “Judgement of Solomon” Help the Open Source Community Find Success? | Michele Dolfi | 23m |
OSI produces a concrete definition that requires access to model weights, training code, dataset provenance (when legally shareable) and claims those freedoms… | ||
| Meet Docling: The “Pandas” for Document AI | Peter Willem Jan Staar, Cesar Berrospi Ramis | 01h 07m |
Docling is an open‑source Python package that converts PDFs, DOCX, PPTX, HTML, images and Markdown into a structured DoclingDocument and exports to Markdown or… | ||
| AI/ML Networking Challenges: The Fast and the Finicky! | Lerna Ekmekcioglu | 21m |
Demoed network visibility using a probe‑mesh that measures one‑way delays at Q‑pair granularity, showing a healthy baseline at 1.47 Tbps aggregate with 100… | ||
| Benchmarking GenAI Like a Pro: Scaling Experiments, Predicting Performance, and Keeping Your Sanity | Michael Johnston | 29m |
A reusable benchmarking stack is built and open-sourced as ADO (github.com/ibm/ADO) that runs tens of thousands of GenAI experiments across models, hardware… | ||
| Architecting the AI Bridge: Integrating LLMs and DMN With Drools, Spring Boot, and the MCP Protocol | Tim Wuthenow, Alex Porcelli | 29m |
Architecture that pairs LLMs with executable decision models and a rule engine, demonstrating a live Spring Boot MCP server that serves Drools DMN via MCP, a… | ||
| Monitoring GenAI Applications | Prasad Mujumdar | 22m |
Monocle is an open source Linux Foundation AI&Data project that instruments GenAI apps with OpenTelemetry‑compatible traces, offering Python and TypeScript… | ||
| Ryoma: Building Open Source AI Agents for Enterprise Data Analysis | Hao Xu, Fang Feng | 30m |
Ryoma is an open source AI agent framework that translates natural language into SQL, runs safety checks and prompt validation before calling models, and… | ||
| AI Showdown: Open Source Tools for LLM Face-Offs | Jigyasa Grover, Rishabh Misra | 26m |
An open source workflow called LLM Comparator provides a Python package (pip installable) plus an interactive UI that runs side-by-side comparisons, produces a… | ||
| State of Open Video Generation Models | Sayak Paul | 24m |
Open video models already produce near-closed-model quality: the open 1.2.2/12.2 family rivals Sora on visual fidelity and instruction adherence. Popular… | ||
| Machine Learning on Microcontrollers; Tiny, Useful and Fun | Jon Nordby | 25m |
TinyML systems run inference on-device at under 1 W and often with under 1 MB RAM and a typical bill of materials under $10; emlearn exports scikit‑learn and… | ||
| Who Let the Bots Out? A Guide to Evaluating AI Agents | James Cha-Earley | 25m |
Claims an open source, reproducible evaluation pipeline that splits agent quality into tool use, trajectory (workflow) evaluation, and goal evaluation… | ||
| One Pipeline to Rule Them All: Unifying Multimodal and AI Data Processing with Daft | Sammy Sidhu, Colin Ho | 26m |
Daft is a Python-native, Rust-executed data engine that runs multimodal ETL, document parsing with Dockling, embedding via Hugging Face or OpenAI, and… | ||
| Improve AI Inference (serving models) With KServe and VLLM | Matteo Combi | 34m |
KServe and vLLM combine to serve LLMs on Kubernetes with concrete wins: KServe provides serverless scale-to-zero and multi-model packing, vLLM adds compression… | ||
| The Self-Healing Pipeline: How Dagger + LLMs Are Making Traditional CI/CD Obsolete | Vikram Vaswani | 42m |
Claim: combining Dagger’s programmable container engine with LLMs produces a self-healing pipeline that diagnoses failing tests, edits code, reruns tests and… | ||
| Organize AI Chaos With Platform Engineering | Sébastien Blanc | 26m |
Platform engineering centralizes LLM metadata (owner, infra region, token budget, status) and exposes opinionated self‑service for RAG stacks by modeling… | ||
| From Deep Research to Evaluation and Optimization of Advanced AI Age... | Arthur Camara, Jakub Zavrel | 32m |
A modular Deep Research agent is built with planner extractor merger writer stages that call hybrid search (BM25 + vectors + rerankers) and RAG on private… | ||
| Unlocking Insights From Multimodal PDFs Using OpenSearch and Vision-Language Models | Mingshi Liu, Praveen Mohan Prasad | 33m |
Shows two concrete pipelines: one converts PDF tables/images into text (CSV, captions) and stores Titan embeddings via OpenSearch ingest pipelines; the other… | ||
| Technical Workshop: Sarcastically Speaking: Unlocking Multi-modal Sentiment Analysis With NLP and Facial Expressions | David vonThenen | 59m |
Workshop builds a full pipeline from MP4 -> frames -> CSV landmarks and MP4 -> WAV -> acoustic CSV, uses the Mustard dataset with heavy cleaning, dlib 68-point… | ||
| Lightning Talks: Mozilla.ai Blueprints & any-suite - Empowering Devs to Build with Open-Source AI | Stefan French | 16m |
Blueprits and any-suite deliver concrete developer tools: Blueprints templates are runnable prototype apps (hosted demos, GitHub repos) that prioritize small… | ||
| Lightning Talk: Ramalama - The Project That Brings AI and Containers Together While Making AI Boring | Alexon Ferreira de Oliveira | 12m |
Ramalama is a CLI that wraps Podman to run LLMs and vision models as OCI containers, auto-detecting GPUs (CUDA, ROCm, Vulkan, Intel, Ascend) and pulling… | ||
| Running Intelligence at the Edge: Web AI in Action | Muskan Jain | 17m |
Shows a live flow converting PyTorch/TensorFlow models to ONNX, applying quantization/pruning, and serving a 2–4 billion parameter LLM in-browser with weights… | ||
| nothing matches | ||