Allston M.

Plenty of fun and valuable insights to pick through in the Stripe annual letter. Fellow engineers might be most interested to read about how we deploy AI models that learn and adapt in real-time (there is actually a 100 millisecond delay), so every new payment on Stripe learns from the last, and we keep on top of fraudulent actors evolving their tactics. Real the full letter: https://lnkd.in/giy4jgzQ.

556

This is a great writeup that summarizes the different storage approaches used by Datadog over the years and then goes into detail about the newest (v6) engine. LSM and timeseries go well together. https://lnkd.in/g8yTkxuA

97

1 Comment

Great piece by Emilia David on the LangChain open source integration ecosystem, the power of LangGraph, and our newly released LangGraph Platform Helps put everything you see us working in perspective Link: https://lnkd.in/ghajS7zx

100

5 Comments

The Stacklok team has been spending a lot of time looking at ways to improve the security of code generation workflows. Using a modern frontier model helps, but that isn't for everyone or every task. For folks using mid-sized models for simple generation workflows, check out our findings comparing popular models here: https://lnkd.in/gC4PknJY tl;dr; * All models struggle with bad package detection (Llama 3.2-3B was better with 29% packages flagged, but most only came in detecting around 5%). * CVE knowledge is low (with peak awareness of 18% for Qwen2.5-Coder-3B-Instruct, but 5% being more typical) * Insecure code recognition is a mixed bag. * Size isn't everything (some smaller models out-performed larger models) * Newer isn't always better. Check out the linked post for details, and check out Codegate if you are using smaller models for agentic workflows and want extra security.

39

We are excited to announce major updates to RayData. Ray data now has native join support, improved aggregation APIs, and a brand-new shuffle engine. These updates large-scale workloads faster and more efficient while also unlocking lot more capabilities. https://lnkd.in/gRcZDDcp

74

At Ubicloud, we now have data for three generations of hardware. CPU benchmarks give a partial picture. We're sharing how actual workloads perform in alignment with our open source ethos. We looked at GitHub Actions job completion times from 350K CI/CD runs. Our first bare metal instances were Hetzner’s AX 161s. We rolled out AX 162s last year and recently enabled AX 102s for GitHub premium runners. Here’s how their median job completion times compare. * AX 161 (EPYC 7502P) - 153s, baseline * AX 162 (EPYC 9454P) - 129s , 1.19x of baseline * AX 102 (Ryzen 9 7950X3D) - 80s, 1.91x of baseline In other words, your GitHub Actions jobs last year automatically sped up by 1.19x with our fleet upgrade. Faster jobs, lower bills. We recently enabled premium runners that almost double job performance compared to AX 161s. Since you pay by the minute for each completed job, your bills also shrink. In short, using newer hardware matters! If you have any questions or feedback, please drop a comment below. ⬇️⬇️⬇️

36

Juliette Lehmann from my team has done a great job with this experiment, building vector search using Zilliz Vector DB and UltiHash object storage. If you don’t have time to read the full article, here’s a TL;DR 🔹 Why object storage matters for vector search – Efficient retrieval + scalable storage are essential when working with high-dimensional embeddings. 🔹 Experimental setup – Juliette runs vector search on Zilliz, creates embeddings with CLIP, stores raw data in UltiHash object storage instead of local disk, using FlaskAPI for the retrieval pipeline. 🔹 Performance insights – Object storage delivers scalable, cost-effective vector search, but caching and I/O optimizations are crucial for low-latency queries. 🔹 Developer takeaways – Choosing the right storage backend can make or break vector search performance at scale. Object storage isn’t just for cold data—it’s the primary storage of AI infrastructure. It’s a solid deep dive with a step-by-step guide to recreate the setup. For anyone working with vector databases, embeddings, or AI-powered search it's definitely worth a read! Check it out here: https://lnkd.in/dQSm9JdR

25

1 Comment