Berlin Buzzwords 2026
2026-06-07..2026-06-09
-- online --
[Berlin Buzzwords 2012, Berlin Buzzwords 2015, Berlin Buzzwords 2016, Berlin Buzzwords 2018, Berlin Buzzwords 2019, Berlin Buzzwords 2020, Berlin Buzzwords 2021, Berlin Buzzwords 2022, Berlin Buzzwords 2023, Berlin Buzzwords 2024, Berlin Buzzwords 2025, Berlin Buzzwords 2026]
as the organisers put it
BERLIN BUZZWORDS 7-9 June 2026 at KulturbrauereiBerlin Buzzwords is Europe's leading conference for modern data infrastructure, search and machine learning and is focused on open source software projects. BERLIN BUZZWORDS 7-9 June 2026Kulturbrauerei Berlin Buzzwords is Europe's leading conference for modern data infrastructure, search and machine learning and is focused on open source software projects. [...]
| title | author | length |
|---|---|---|
| Building Resilience: The Next Decade of Open Source | Ruth Suehle | 42m |
Open source has become the internet’s backbone but now faces a resilience crisis: Log4Shell (late 2021) forces global regulators to act, many critical… | ||
| How to Tell If Your Agent Used the Right Stuff | Apurva Misra | 41m |
Claim: most agent "failures" are actually failures of context delivery, not model capability. Workflow: instrument traces, save rewritten queries, retrieval… | ||
| AI in the physical world: from observation to discovery | Dmitriy Kostunin, Julian von Hoerschelmann-Schliwinski | 20m |
Next‑generation gamma‑ray array spans 50 telescopes in two hemispheres, producing 10 ns camera exposures and about 6 PB/year of raw data; engineers use deep… | ||
| What you should know about constraints in PostgreSQL 18 | Gülçin Yıldırım Jelínek | 41m |
PostgreSQL 18 adds concrete changes to constraints: temporal keys (primary/unique/foreign keys over range periods, implemented via btree_gist/EXCLUDE-style… | ||
| Detecting Hidden Bias in Datasets Before Models Fail | Stas Don | 30m |
A language‑learning churn model shows strong offline numbers — 80% precision — yet production recall falls by about 40% recall drop because analytics only… | ||
| What If We've Been Scaling Stream Processing Wrong All Along | Hartmut Armbruster | 41m |
Argues that default horizontal distribution is often needless and builds a single-node alternative preserving Kafka Streams DSL, implementing Flink-like… | ||
| DuckDB beyond the notebook | Matthias Niehoff | 40m |
DuckDB is shown running in the browser via DuckDB browser WebAssembly, loading Parquet/CSV from S3 or local files and connecting to Arrow/Polars without… | ||
| Circular Dependency Fixes when Bootstrapping a Golden Set | Radu Gheorghe, Rafał Kuć | 20m |
Workflow for bootstrapping a golden set that actually works: start from a 62K MovieLens sample enriched via TMDB, index into Vespa + navigator + Python… | ||
| One GPU, Four Retrieval Modes: Multi-Model Search Serving | Filip Makraduli | 42m |
Shows how one server process runs dense, sparse, ColBERT and cross‑encoder retrieval together on a single GPU by exposing three primitives: encode, score… | ||
| Zero downtime index upgrade in Apache Solr | Rahul Goswami | 42m |
Apache Solr now offers a zero-downtime in-place upgrade path (coming in 9.11 with Lucene changes exposed in 10.1/10.4) that converts older segments by… | ||
| Empowering OSS maintainers in the age of AI | Jarek Potiuk | 42m |
Apache Airflow sees a jump from ~100 to ~500 open PRs and security reports rise from ~10 to 30+ per month, so a skill‑based agent stack triages, labels and… | ||
| The Three-Body Problem of Inverse Hybrid Search | Ravindra Harige | 40m |
Solves saved-image alerts by turning product arrivals into queries against millions of stored searches, combining vector+boolean+lexical matching with strict… | ||
| Beyond Grep: Search for Reliable Coding Agents | Amine Gani, Roudy Khoury | 36m |
Claim: improving retrieval beats blindly enlarging models. Compares grep, lexical ranking, structural search (function/class extraction + LSP-like links) and… | ||
| Correctness Too Cheap To Meter: Formal Verification and LLMs | Emilie Ma | 33m |
Argues that LLMs plus formal tooling can automate spec creation: a pipeline called Specula mines repos, generates TLA+ models, instruments code, runs TLC model… | ||
| From Legacy Search to Vespa: What a Real PoC Taught Us | André Charton, Valeriia Platonova | 20m |
Germany’s largest classifieds platform runs 60 million items with ~1M new items/edits/deletes per day and moves the Motors homepage feed from Elastic to Vespa… | ||
| Kafi Streams: Complex Stream Processing Made Simple | Ralph Matthias Debusmann | 41m |
Kafi Streams is an open source Python library built on PyDBSP that treats streams as weighted sets so streaming behaves like batch: strong, deterministic… | ||
| Escaping the Cloud: High-Performance AI in your Browser | Johannes Kolbe | 41m |
Argues that full inference can move into the browser using WebGPU, ONNX Runtime and Transformers.js, and demonstrates LLMs, speech and vision running locally… | ||
| Why Choose One: Multi-Engine Analytics with Apache Wayang | Zoi Kaoudi, Haralampos Gavriilidis | 42m |
Apache Wayang decouples platform-agnostic plans from execution and maps Java/Scala/Python/SQL operators to Spark, Flink, JDBC and ML backends using an inflator… | ||
| Constant-Time Aggregations with Star-Tree in OpenSearch | Shailesh Kumar Singh | 41m |
OpenSearch adds a Star-Tree index that materializes multi‑field aggregates inside segments by extending Lucene DocValues to store sums, counts and sketches… | ||
| Low-Resource Languages as Stress Tests for NLP Data | Priscilla Lola Adenuga | 17m |
Low-resource languages expose where annotation and evaluation collapse real ambiguity into a single label: examples like old friend and Oge’s surface form E ho… | ||
| C++ Search for Database Kernels: Built In, Not Bolted On | Andrey Abramov | 27m |
IResearch is an Apache 2.0 C++ library built to embed inside database kernels with WAL integration and transactional consistency and it benchmarks against… | ||
| The Failures That Don't Crash: MLOps for AI Agents | Bartosz Mikulski | 18m |
Four concrete reliability patterns for production agents: shadow test against a baseline by comparing distributions (latency, first-token time, token cost… | ||
| Personalize Search Results with OpenSearch Agentic Memory | Hajer Bouafif | 21m |
Claims a working pipeline that personalizes identical queries by having lightweight agents read and write OpenSearch agentic memory (released in 3.3) and… | ||
| Tensor arithmetics in search and ranking for Ecommerce. | Piotr Kobziakowski | 22m |
Small 256-d fashion vision model plus Vespa’s ranking engine runs live tensor arithmetic at query time to steer results by color, shape, style and text facets… | ||
| Let LLMs Wander: Engineering RL Environments | Stefano Fiorucci | 39m |
Builds a Tic-Tac-Toe RL environment with the open-source Verifiers library, using deterministic Python verifiers, XML parsing for moves, and rewards of 1/0.5/0… | ||
| 10x CouchDB Performance Gains for a AAA Game Launch | Jan Lehnardt | 42m |
Benchmarks using Apache Bench and iPerf reveal a 1 Gbps AWS networking cap and huge latency spikes for concurrent view queries. Changes: enable higher AWS… | ||
| Scientific Data Under Threat in Today’s America | Uwe Schindler | 22m |
Federal research grants drop 20–30% in 2025, staff cuts lead NOAA to decommission and stop refreshing long‑running climate and biodiversity series, and many… | ||
| Ultraviolet: Turn Hidden Document Data into an AI Advantage | Alessio Vertemati | 20m |
PDF internals contain hidden layers that alter AI pipelines: invisible text (opacity, tiny fonts, cropped or behind images), Unicode tricks including 15K… | ||
| Apache Solr 10: What's Coming up for Vector Search | Alessandro Benedetti, Ilaria Petreti, Anna Ruggero | 41m |
Solr 10/10.1 brings concrete vector-search plumbing: scalar quantization (compress float32 to 4 or 7 bits) and binary quantization (index as 1‑bit vectors… | ||
| Towards Chunk-less RAG | Carles Onielfa | 18m |
Extracts token vectors from Qwen3-Embedding-0.6B and computes cosine between every query and document token to produce relevance heatmaps, clusters… | ||
| GitOps for n8n: Treating Workflows as Code | Joao Gilberto Magalhaes | 47m |
A Python CLI called n8n-gitops uses the n8n REST API to export workflows into a structured repo (YAML manifest, workflows JSON, and a scripts folder that… | ||
| Keeping data private in real-time pipelines | Olena Kutsenko | 40m |
Covers concrete patterns for streaming privacy using Apache Kafka, Apache Flink, and Apache Iceberg: masking is trivial but irreversible, tokenization requires… | ||
| Reviving phonetic algorithms for better search relevance | Pietro Mele, Radu Pop | 22m |
Reimplements century‑old phonetic encoders and benchmarks them against Levenshtein fuzziness on a huge French audiovisual index, finding phonetic matching cuts… | ||
| Floe: Policy-Based Table Maintenance for Apache Iceberg | Neelesh Salian | 43m |
Open-source Floe turns Iceberg maintenance into declarative policies that run via Spark or Trino and connect to 7 catalogs (REST, Polaris, Lakekeeper… | ||
| AI is here – time to throw away our search engines? | Charlie Hull, Atita Arora, Jo Kristian Bergum, Dmitry Kan, Evgeniya Sukhodolskaya | 41m |
Claims that AI simply replaces search get tested with concrete techniques: use vector embeddings for multimodal data, apply BM25 + rerank by computing… | ||
| Herding Cat(alogue)s: The Apache Iceberg™ Catalog Landscape | Danica Fine | 38m |
Concrete comparison of Iceberg catalogs: HMS uses database locks and Thrift so it’s ubiquitous but coarse‑grained; AWS Glue uses conditional writes and a… | ||
| OpenSearch Software Foundation: 1 Year of Open Governance | Kris Freedain, Carlos Rolo | 20m |
Project migrates into the Linux Foundation technical and governance stack and establishes TAGs, SIGs, a technical steering committee, an LTS program and… | ||
| Agentic Retrieval: Building Self-Optimizing Search Systems | Jo Kristian Bergum | 40m |
Claims that agents compress relevance feedback from months to seconds and can tune scoring, schema and indexing in real time, demonstrated by Hornet’s… | ||
| No 0-day required, just target the AI coding assistant! | Leo Visser | 40m |
Demos show AI coding assistants accepting hidden instructions (PowerShell prompt with overlapping characters that hides “UCP25” and makes an item free), white… | ||
| OSS Security: Lessons from 10+ Years at Apache Solr | Jason Gerlowski | 37m |
Timeline of concrete choices and outcomes: Solr adds pluggable auth added in 2015 (Kerberos, Hadoop, JWT/OAuth) instead of reusing Shiro, which increases… | ||
| Apache Spark Declarative Pipelines in Action | Frank Munz | 42m |
Spark 4.1 Declarative Pipelines replace imperative glue code with declarative configs and runs pipelines that auto-handle validation, dependencies, retries and… | ||
| Text-to-Struct: Fine-tuning SLMs for Query Intent | Hugo Jimenez, Sandra Bullón | 42m |
Fine-tuning a 4B SLM (Gwen) via LoRA/QLoRA and knowledge distillation from a Sonnet 4.6 annotator turns raw user or agent prompts into Structured Semantic… | ||
| How Apache Iceberg Enables Multi-Engine Data Platforms | Geetha Anne | 31m |
Apache Iceberg decouples storage from compute so Spark, Trino and Flink read and write the same table using metadata-driven concurrency and snapshot-isolation… | ||
| From OLTP to OLAP: Is PostgreSQL Eating Analytics Too? | Daniel Seybold | 19m |
Postgres with columnar extensions such as pg_duckdb, Citus and TigerData runs on EC2 and is compared to ClickHouse Cloud using ClickBench (104-column flat… | ||
| Building a Local News RAG: The Quest for Trustworthiness | Marcel Dokters | 35m |
A small five-person team (data scientist, cloud engineer, product manager, journalist, UI/UX) builds a local-news RAG for a 1.5-year archive (~500k articles)… | ||
| SPRUCE it up! Open Source GreenOps at scale | Julien Nioche | 39m |
SPRUCE is an Apache-licensed pipeline that actually enriches cloud usage reports (AWS, Azure) with added columns for energy, CO2 and water and writes enriched… | ||
| Observability’s Sixth Sense: Detecting Anomalies in Metrics | Virginia Diana Todea | 35m |
Shows a live workflow that wires OpenTelemetry into VictoriaMetrics queries and a backend that runs MAD, Z‑score and Prophet models, plus an MCP server that… | ||
| Beyond the Hype: When Apache Flink Solves Real Problems | Naci Simsek | 41m |
When Flink becomes essential: event-driven fraud detection, CDC that syncs RDBMS to search in sub-second, and low-latency dashboards where freshness matters… | ||
| The Agent Era: How AI Agents Are Reshaping Data Platforms | Monica Sarbu, Danica Fine, Philipp Krenn, Till Döhmen | 47m |
Agents hit platforms with dozens of parallel queries, create ephemeral databases constantly, and break assumptions about human review and rare DB creation: a… | ||
| Dynamic Broker-Side Filtering for Kafka | David Kjerrumgaard, Álvaro Rodríguez | 40m |
Claims a broker-side filtering implementation that closes a 7‑year KIP gap by shipping a Kafka‑compatible filtering layer (native Java/Python SDKs, control… | ||
| Context-Aware Segments: Solving the "Scatter-Read" Problem | Rishav Sagar, Tejas Shah | 41m |
Introduces Context-Aware Segments (OpenSearch RFC #18576, Lucene Issue #13387) that enforce document locality by routing documents to group-specific writers… | ||
| Mentoring In Open Source in the Age of AI | Tilda Udufo, Busayo Ojo | 40m |
Open source mentorship with Outreachy shows that contributors submit polished, AI‑generated code they cannot explain, causing maintainers to waste review time… | ||
| Streamling: Lightweight, Extensible Streaming on DataFusion | Xiao Meng, Rafael Aguiar | 40m |
A Rust streaming engine built on Apache DataFusion and Arrow turns Kafka Avro into continuous Arrow RecordBatch streams using DataFusion’s planner and custom… | ||
| Relevance Feedback Inside the Search Engine | Evgeniya Sukhodolskaya | 21m |
A new API injects model-provided relevance scores directly into the vector index so the HNSW traversal picks next hops by a weighted mix of cosine similarity… | ||
| Turning the database inside out again | Tom Scott, Roman Kolesnev | 42m |
Claim: combine Apache Kafka and Apache Iceberg so the event log is the canonical store, using an Iceberg proxy for Kafka that serves Kafka offsets as… | ||
| Search is Back: Solving the "Context Crisis" for AI Agents | David Louis Hollembaek, Vincent Pistor | 38m |
Shows how to combine Knowledge Graphs + Vector Search with open source tooling (Cogni, Quadrant, TurboPuffer) to give agents structured context, not just… | ||
| Event-driven Agents with Complex Event Processing in Flink | Steffen Hoellinger | 43m |
Claims a concrete architecture: use Apache Flink CEP to detect sequences and filter millions‑per‑hour Kafka events, run statistical anomaly detection or simple… | ||
| When better retrieval makes agents worse | Lester Solbakken | 37m |
Argues that improving retrieval scores can worsen looped agents because hard distractors—plausible but wrong passages like “2022” vs “2026”—steal attention… | ||
| Writes, 3 ways: Postgres, Apache Kafka® and Apache Iceberg™ | Celeste Horgan | 20m |
Compares how writes actually happen: Postgres write-ahead log records every SQL change and enforces atomic, disk-first transactions so correctness beats raw… | ||
| How to Survive the Vortex of LLM Change | Carmen Iniesta, Carles Onielfa | 20m |
Progress Semantic Rack is a production intelligent search system that adapts by running RAG and a small LLM as an internal evaluator that scores every user… | ||
| Sunset for the Wild West: Making ML disciplined by default | William Benton | 39m |
Claims that dependency chaos makes ML systems insecure and unreproducible, supported by a PyTorch container dependency graph with ~1,500 nodes and ~6,000… | ||
| Real-Time ML Pipelines: Feature Chaining with Chronon | Varant Zanoyan | 36m |
Chronon (Zipline) is an open‑source feature+embedding platform that unifies batch and stream computation to serve sub‑100ms features and embeddings for… | ||
| From Inverted Index to Columnar Vectorized Execution Search | Ankit Jain | 37m |
Shows how Lucene/OpenSearch moves from per-document scoring to batch columnar paths: BKD range counting cuts an 8s monthly histogram to ~200–300ms (50x… | ||
| nothing matches | ||