We're excited to be back with another evening Database Internals!
This time, we’ll explore how modern data platforms are moving beyond batch-oriented architectures toward fresh, streaming-first, low-latency systems. With talks on Flink, Fluss, Kafka, and real-time context for agents, the evening will look at where data architecture is heading next.
As always, there will be plenty of time for networking, discussion, and snacks 🍕.
Event Details:
📍 Venue: Mindspace, Herzogspitalstraße 24
📅 Date & Time: Wednesday, June 10 2026, doors open at 18:30
- Realtime LLM pipelines in FlinkSQL
Abstract: In this session, Joseph shows how Ververica's enterprise-ready Apache Flink can score incoming events live against a large language model — turning a raw stream into an enriched, decision-ready one in flight. The demo runs a continuous stream of customer reviews through a single declarative Flink SQL job: each review is classified the moment it arrives, tagged with a sentiment label, and written straight to Apache Kafka as enriched JSON, with per-record latency captured along the way.
The heart of it is Flink's ML_PREDICT, which treats the LLM as a first-class SQL operator — a streaming source feeds each comment to the model, which returns one of positive, negative, neutral, or mixed, and the result lands in a Kafka topic. No glue code, no separate serving stack: just source, model, and sink in plain SQL, running on a platform built for production.
Speaker Bio: Joseph Gade is a Sales Engineer at Ververica, the original creators of Apache Flink. He brings 20 years of experience in the data space, focusing mainly on real-time technologies and analytics.
- Realtime Context Engine - a realtime analytical database
Abstract: As LLM applications move from chatbots to agents, they need a standard way to fetch operational data instead of relying only on RAG or periodically synced databases. MCP is becoming that interface, but it does not solve the underlying serving problem: turning streaming data into fresh, low-latency, governed context that an agent can query reliably. Real-Time Context Engine (RTCE) sits in that layer by materializing state from Kafka streams and exposing it through MCP. Lightning Tables is the new analytics database that backs RTCE. It is the Kafka-native serving and analytics engine built to query live data from Kafka and Iceberg tables via Tableflow using an Arrow in-memory layout, vectorized execution, and locality-aware scheduling. In this talk, we will walk through the architecture and the performance tradeoffs around freshness, point lookups, scans, and concurrency.
Speaker Bio: Ryan Murray is a Director at Confluent, leading the Lightning Table and Tableflow teams. He is a founder, an Apache Iceberg committer, and has done everything from database engineering to bond trading to theoretical physics. Ryan still dreams of winning the Stanley Cup one day.
Agenda:
🔹 18:30: Doors Open
🔹 18:40: Welcome
🔹 18:45: Talk #1: Ben Gamble (Ververica)
🔹 19:30: Pizza & Networking 🍕
🔹 20:00: Talk #2: Ryan Murray (Confluent)
