talk-data.com
People (6 results)
See all 6 →
Activities & events
| Title & Speakers | Event |
|---|---|
|
We Hacked Ourselves So You Don't Have To: An Elasticsearch Battle Story
2025-11-12 · 19:00
Maryna Kryvko
– Senior Software Engineer
@ Zalando
,
Ivan Potapov
– Research Engineer
@ Zalando
This talk examines the October 2024 Zalando outage, which resulted in severe revenue impact. The incident was triggered by an automated security scan that uncovered an unprotected GraphQL endpoint. This led to an unexpected amplification of requests, causing computationally expensive queries against the service in front of the search indices and ultimately overloading the Elasticsearch clusters. The prolonged impact highlighted difficulties in pinpointing the root cause of high load in Elasticsearch, and proved that even though a perfect storm is by definition a rare occurrence, it still should never be discarded. Sometimes, when you hear the hoofbeats, they are zebras after all. |
|
|
The Performance Paradox: When Slower Code Doesn't Slow You Down
2025-11-12 · 18:30
Anderson Queiroz
– Senior Software Engineer
@ Elastic
Abstract: We added on-the-fly gzip decompression to Elastic Filebeat and the Elastic Agent—our log collection tools—to enable the ingestion of gzip archives and rotated logs. A performance drop was expected, so we benchmarked the feature only to find that the performance didn't drop at all. This talk is the story of our hunt for a non-existent bottleneck and how a holistic view of application performance uncovered the surprising truth about where the real costs lie. |
|
|
Practical guide: Fine-tuning Qwen3 with LoRA
2025-10-30 · 11:30
Ivan Potapov - Research Engineer, Zalando SE In this workshop, we fine-tune Qwen models with parameter-efficient adapters using two complementary approaches: Soft Prompt token tuning and LoRA SFT, with an optional KL-anchored SFT term to keep the model’s behavior close to the base while adding new styles and formats. You’ll see how to prepare open-source data (Dolly 15k), render with chat templates, run short training loops, and monitor validation loss/perplexity with stepwise evaluations. A tiny KL toy example explains per‑token contributions to H(P)\, H(P\,Q)\, and KL(P\|\|Q)\, making the “anchoring” intuition concrete. By the end\, you’ll know how to apply Soft Prompt for quick style steering\, LoRA for deeper adaptation\, and KL regularization to reduce drift and forgetting—plus how to save/load LoRA adapters for deployment. About the speaker: Ivan Potapov is a Research Engineer at Zalando, specializing in search. He has taught workshops on data engineering, AI agents, and LLM alignment, helping practitioners bridge software engineering with applied machine learning. Join our slack: https://datatalks.club/slack.html |
Practical guide: Fine-tuning Qwen3 with LoRA
|