Berlin Buzzwords 2022
2022-06-12..2022-06-14
-- 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]
| title | author | length |
|---|---|---|
| Meet the people fighting surveillance capitalism | Fiona Coath | 33m |
Looks at concrete fights and tactics: shows Cambridge Analytica using 50 million Facebook profiles for personality-targeted political ads, whistleblowers… | ||
| Hybrid search: Greater than the sum of its parts? | Lester Solbakken | 45m |
Shows that combining BM25 and transformer embeddings in Vespa with an HNSW index and sentence-transformers (miniLM) yields better retrieval than either alone… | ||
| AI-powered Semantic Search; A story of broken promises? | Jo Kristian Bergum | 39m |
Semantic vector search often fails out of domain: models trained on MS MARCO: 9M passages, 500k labels, 3,600 test queries give great in-domain results but… | ||
| Neural Search Comes to Apache Solr: Approximate Nearest Neighbor, BERT & more | Alessandro Benedetti | 47m |
Apache Solr adds first-class neural search in 9.0: store vectors in a dense vector field, index with Lucene’s HNSW proximity graphs and run Approximate KNN… | ||
| What's new in Apache Solr 9.0 | Anshum Gupta | 43m |
Solr 9.0 adds Lucene 9 dependency and a real feature set: support for dense vector field search and new language stemmers, experimental rate limiting and task… | ||
| Understanding Vespa with a Lucene mindset | Atita Arora | 20m |
Vespa is a low latency platform that combines real time ingestion, search and analytics with built‑in ML and TensorFlow support and true partial updates so… | ||
| Cross-Platform Data Lineage with OpenLineage | Julien Le Dem | 46m |
OpenLineage defines a lightweight event model that captures start/complete/fail events for each run, attaching extensible facets with JSON schemas (schema… | ||
| Muves: Multimodal & multilingual vector search w/ Hardware Acceleration | Aarne Talman, Dmitry Kan | 37m |
Muves combines Elasticsearch/OpenSearch with a GSI APU backend to run multimodal multilingual vector search using CLIP and multilingual sentence transformers… | ||
| Change data capture with Debezium…and without | Petros Angelatos | 19m |
Materialize implements a direct Postgres CDC connector that skips Kafka+Debezium and reads Postgres logical replication, decoding the WAL into commit-ordered… | ||
| Patterns and anti-patterns for production ready Kafka Streams apps | Christoph Schubert | 41m |
Covers concrete, actionable patterns: use the Topology Test Driver and Testcontainers for unit and integration tests; prefer mapValues/flatMapValues so key… | ||
| Changelog Stream Processing with Apache Flink | Timo Walther | 39m |
Apache Flink SQL processes tables as change logs and ships an optimizer that turns streams into Flink SQL as changelog processor semantics: four change kinds… | ||
| Word2Vec model to generate synonyms on the fly in Apache Lucene | Daniele Antuzi, Ilaria Petreti | 38m |
Implemented an on-the-fly Word2Vec synonym filter for Apache Lucene that reads a trained deep learning4j model, builds an HNSW proximity graph from embeddings… | ||
| Running Apache Spark on K8s: From AWS EMR to K8s | Ramiro Alvarez Fernandez, Álvaro Panizo, Daniel Hernández Alfageme | 41m |
Spark is a trend technology that it is being used for a lot of companies for large-scale data analytics. During the first try, companies usually try to use the… | ||
| Architecting Solr indexing pipelines in Google Cloud Platform | Shubhro Jyoti Roy | 34m |
Box moves its Solr pipeline off HBase into GCP using Bigtable, Dataproc, Pub/Sub and Cloud KMS, while keeping Solr on-prem and running dual writes to avoid… | ||
| Using Solr unconventionally to serve 26bn+ documents | Richard Goodman | 30m |
Brandwatch rearchitects Solr on AWS to serve 26bn+ documents by splitting writes and reads: a compact write cluster snapshots indexes to S3, and multiple… | ||
| Do It Yourself: Programmable Metrics using OpenTelemetry | Ricardo Ferreira | 44m |
Builds a Java Spring Boot microservice (repo branch b buzz) and wires OpenTelemetry end-to-end: local agent + OpenTelemetry Collector receiving OTLP, exporter… | ||
| Kafka Monitoring: What Matters! | Amrit Sarkar | 36m |
Describes concrete monitoring: export JMX → Prometheus → Grafana for broker metrics and use Burrow lag detection for consumer offsets; alert on four definite… | ||
| Offline Ranking Validation - Predicting A/B Test Results | Andrea Schütt, Yunus Lutz | 24m |
Otto's search team builds an offline validator that predicts A/B outcomes by reweighting clicks with inverse propensity scoring and by correlating model NDCG… | ||
| NrtSearch: Yelp’s fast, scalable, and cost-effective open source search engine | Umesh Dangat | 34m |
Yelp builds NrtSearch a Lucene-based engine that uses Lucene segment replication and a primary‑only writer that uploads segments to S3 while replicas pull… | ||
| Building an Open-source Framework for Generating Embedding Vectors | Frank Liu | 40m |
An open-source framework called towhee provides operator-based pipelines that glue together models, ETL, and vector storage so teams can prototype embedding… | ||
| Effective CI/CD for Large Systems | Josh Reed | 43m |
Argues CI must be fast and component-driven: targets 5–15 minute CI for quick feedback and 15–30 minute deployments for safe rollbacks, else developers batch… | ||
| URL Frontier, an open source API and implementation for crawl frontiers | Julien Nioche | 21m |
An open source crawl-frontier with a gRPC/protobuf API and Java client, queuing by key (usually host), per-URL scheduling dates, streaming get/put methods… | ||
| Next generation OLAP stack using Apache Pinot | Chinmay Soman | 34m |
Apache Pinot is a distributed columnar OLAP engine that claims real-time user-facing analytics at scale: ingesting millions of events/sec and serving over… | ||
| Goodbye Tracking, Hello Privacy | Nina Müller, Lara Menéndez García | 42m |
Claim: web users are tracked hundreds of times daily — 376 times per day in Germany and 747 in the US — and GDPR/CCPA often fail to stop it. Built front-end… | ||
| Apache Kafka simply explained | Olena Kutsenko | 41m |
Explains that Apache Kafka is a distributed, scalable event-streaming platform that handles millions of records per day and stores petabytes persistently… | ||
| Relevance is not a Thing but a Perception | Ana Maria García Sánchez | 26m |
Argues that relevance is not a static score but a user perception and shows working prototypes that replace rank lists with motion and correlated visual action… | ||
| What we learned from reading 100+ Kubernetes Post-Mortems | Noaa Barki | 30m |
Team reads 100+ post-mortems and finds recurring root causes: bad cronjob concurrency settings that spawn thousands of pods (Target, Zalando), missing memory… | ||
| Matscholar: The search engine for materials science researchers | John Dagdelen | 24m |
Search engine indexes text from 5 million abstracts to extract millions of properties using a mix of supervised NLP, bespoke language models and embeddings… | ||
| Live build: How to harness streaming data in real time | Tomas Neubauer, Javier Blanco Cordero | 45m |
Live demo wires a Bluetooth heart‑rate strap -> phone -> websocket gateway -> Kafka, then runs layered microservices that interpolate timestamps, compute… | ||
| The perils of building a democratic data platform | Andre Jasiskis, Joaquim Torres | 37m |
Company runs a CDC ingestion pipeline on Kafka and a centralized processing layer that handles 80000 unique datasets and over several petabytes read per day… | ||
| Cloud-native ETL with Java Quarkus, Kubernetes, and Jib Container Builder | Hakan Lofcali | 30m |
Cloud-native ETL architecture with Java Quarkus, Kafka/Redpanda, and Jib shows a concrete pipeline: GitHub Actions builds artifacts, ArgoCD deploys immutable… | ||
| Compress language models to effective & resource-saving models with knowledge distillation | Qi Wu | 35m |
Distills a Flair pipeline based on XLM-R teacher into an RNN student with GloVe for German NER, training the student on teacher-labeled data plus originals… | ||
| Entity Linking at scale with Lucene | Edoardo Tosca | 39m |
Signal AI replaces regex-heavy rule matching with Apache Lucene using a percolator model and custom analyzer so documents are tokenized once and queried many… | ||
| Neural Search - Let's talk about quality | Florian Hoenicke, Maximilian Werk | 33m |
An open-source finetuner is used with Elasticsearch plus vector search to improve e‑commerce fashion retrieval on a 410‑item WDC corpus; plain CLIP performs… | ||
| Working in the Open...Search | Sean Neumann, Charlotte Henkle | 21m |
OpenSearch forks Elasticsearch 7.10.2 and Kibana 7.10.2 into an Apache 2.0 project and ships OpenSearch engine plus OpenSearch Dashboards. Early work removed… | ||
| Searching through large graphs using Elasticsearch | Radu Pop | 44m |
Project turns a cultural-heritage graph into multiple indices and small flattened JSON fragments stored in MongoDB, transformed with declarative GSLT/JSLT… | ||
| Autoscaling Elasticsearch for Logs on Kubernetes | Radu Gheorghe, Ciprian Hacman | 32m |
Explains a concrete autoscaling design for Elasticsearch/OpenSearch on Kubernetes: rotate time-based indices by size (recommended 10 GB per shard), update… | ||
| Optimizing Containers for Security and Scaling | Thomas Fricke | 35m |
Shows how building minimal images from scratch and a simple hardening script cuts image size and attack surface — nginx drops from 142 MB (Alpine) to 8 MB… | ||
| Solving the knapsack problem with recursive queries and PostgreSQL | Francesco Tisiot | 41m |
Shows how to solve knapsack inside PostgreSQL using WITH RECURSIVE CTE that builds arrays of picked items, tracks total weight and value, and enforces… | ||
| Should we stop using distance in our location-based data recommendation models? | Charlie Davies | 42m |
Argues straight-line distance often misranks results and shows an API + Elasticsearch plugin that computes travel-time ranks from actual transport networks… | ||
| Scaling your Kafka pipeline can be a pain - but it doesn’t have to be | Opher Dubrovsky, Ido Nadler | 33m |
A multi‑petabyte Kafka pipeline (25B events/day ≈300k eps, 5M eps peak, 60 brokers, 9k partitions) moves from continuous microbatch Spark readers to hourly… | ||
| The Race to the Bottom: Low Latency in the age of the Transformer | Max Irwin | 23m |
Shows a practical CPU-first path to transformer inference: a Rust-based Mighty inference server + ONNX Runtime cluster runs 128 inference cores and… | ||
| Scaling an online search engine to thousands of physical stores | Aline Paponaud | 38m |
Describes building an Elasticsearch platform that indexes one index per physical store to keep local offers independent, accepts a marketplace feed of millions… | ||
| Learning about AI/ML for Text, with Wordle! | Nick Burch | 34m |
Uses Wordle as a concrete text problem and shows working Python code that pulls five‑letter words from the Ubuntu dictionary, scores candidates by letter… | ||
| Luxuries, necessities, and the challenges that remain | Sophie Watson, William Benton | 41m |
Shows concrete results using Nvidia hardware and Python: cuPy and RAPIDS let UMAP run at interactive latency so visual exploration uses full data instead of… | ||
| Min and Max Aggregations with Updates in Real Time | Minakshi Korad | 20m |
Real-time ingestion uses a Kafka Streams app to compute count, sum and newly added min/max over roll-ups (minute/hour/day) while supporting record updates; the… | ||
| Logging Apache Spark - How we made it easy | Simona Meriam | 16m |
Nielsen runs Spark on Amazon EMR processing more than 60 terabytes and billions of events daily and the team builds a bootstrap bash script that installs… | ||
| Reproducible and shareable notebooks across a data science team | Pascal Godbillot, Mike Tapi Nzali | 37m |
CybelAngel processes about a billion candidate leaks and is scaling a data team that plans to double; notebooks drive exploration but were hard to share… | ||
| Build Real-time Analytic Applications: The Easy Way | Sergio Ferragut | 40m |
Apache Druid claims sub-second queries on streaming and historical data by building immutable columnar segments, keeping real-time shards in-memory on middle… | ||
| A smooth ride: Online car buying and selling at mobile.de | Ricardo Kawase | 27m |
mobile.de runs a suite of production data products: price prediction using XGBoost/H2O with BigQuery + Airflow and a Scala REST API, a personalized search… | ||
| Don't Panic: Getting Your Infrastructure Drift Under Control | Eran Bibi | 30m |
Claims that 90% of large IaC deployments show drift and that 50% of those drifts go unnoticed, with concrete fixes: run terraform plan with detailed-exitcode… | ||
| The life of a search engine administrator | Lucian Precup, Vincent Bréhin | 43m |
Describes running an e-commerce and enterprise search stack with a custom business console that combines config, logs, pilot, debug and data-modeling modules… | ||
| Scaling the Open Source Climate Community | Erik Erlandson | 36m |
OS-Climate is an open community platform that standardizes ESG data and models to let smaller financial firms do portfolio alignment; it runs on AWS S3… | ||
| Benefits of MQTT for IoT Messaging and Beyond | Mary Grygleski | 44m |
MQTT started as a satellite oil‑pipeline protocol and now targets cloud‑scale IoT with concrete features: QoS 0/1/2 for delivery guarantees, last will and… | ||
| Unpaired Sentiment-to-Sentiment Translation: Using Reinforcement Learning | Sakshi Deo Shukla | 32m |
Produces semi‑supervised parallel corpora from unpaired reviews by running a cycle reinforcement learning loop of a neutralization and emotionalization modules… | ||
| Help! I Need To UnSQLize My Application | Joel Lord | 43m |
Shows how to move a normalized SQL schema into MongoDB by converting joins into embedded arrays or controlled lookups, using an aggregation pipeline to… | ||
| nothing matches | ||