Irene Levina

MCP C# SDK is now v1.0! The release brings full support for the latest MCP spec (2025-11-25), including better auth, richer metadata, tool-calling improvements, safer user interaction patterns, and cleaner support for long-running operations. https://lnkd.in/efZHx9nr #csharp #dotnet #mcp

15

4 Comments

Your CI/CD bill is not exploding because of “expensive runners,” it is bleeding out through flaky tests, over-parallelized jobs, and slow feedback loops that quietly tax every deploy This piece shows you how to redesign pipelines so you cut compute waste, shrink queues, and speed up developer flow without adding yet another approval gate or YAML religion Full breakdown in the comments. #DevOps #CICD #PlatformEngineering

8

1 Comment

Even frontier LLMs can use a helping hand when it comes to detailed API documentation for fast-moving open source libraries and tools. I put together a quick cheatsheet generator, and a repo of cheatsheets (generated by LLM, of course, from the source projects' github docs repos). Example for Bun: https://lnkd.in/gAYCU5bz #bun #llm

15

Human to Claude - read the API at <url>, create a Jay Contract (YAML file description for UI) for the result object. Here are the docs for the contract file. Claude - generated contract that compiles and works as expected (impressive), just with 50% of (missing fields, hallucinated fields, etc. ). Human to Claude - read the API again and compare. Claude - 🤦‍♂️ , then fixed. 100% correct.

20

6 Comments

New blog post, "Symmetric ACP": https://lnkd.in/dDU7FDbq This post describes **SymmACP** -- a proposed extension to Zed Industries' Agent Client Protocol that lets you build AI tools like Unix pipes or browser extensions. Want a better TUI? Found some cool slash commands on GitHub? Prefer a different backend? With SymmACP, you can mix and match these pieces and have them all work together without knowing about each other.

18

Nice blog post from CockroachDB introducing value separation, a technique pioneered by WiscKey (Lu et al., FAST'16), an SSD-optimized LSM-tree design that separates keys (sorted in the LSM) from values (appended to a value log) to minimize write amplification during compactions. Pebble, the key-value storage engine underlying CockroachDB, implements value separation to optimize LSM-tree compactions for workloads with large values by storing values >= 256 bytes in separate blob files and replacing them in SSTables with compact value handles containing metadata (blob file ID, block ID, value ID, and length). This decouples key sorting from value movement, allowing compactions to copy small handles instead of rewriting full values, significantly reducing read/write amplification and I/O bandwidth during standard compactions. Blob files use a columnar-block format with index blocks for constant-time value lookups, while a new blob rewrite compaction type reclaims space by identifying live values via run-length encoded bitmaps and consolidating them, triggered by accumulated dead data or excessive blob reference depth (a metric tracking distinct blob files per SSTable range). Iteration and reads resolve handles by fetching from cached blob file blocks, preserving correctness for MVCC and handling fragmentation through rewrite compactions. Benchmarks show up to ~47% throughput improvement in 4 KiB value upsert workloads, though trade-offs include potential read path indirection overhead and space amplification, mitigated by heuristics and selective separation. Similar Implementations: RocksDB has incorporated similar ideas through its Titan extension (a leveled-compaction variant with blob files for large values) and later integrated blob support, addressing compaction overhead for immutable large blobs. More recently, LavaStore (Wang et al., VLDB'24), ByteDance's custom LSM-based engine for cloud services, employs tailored key-value separation with optimizations like scavenger-based garbage collection. Pebble's implementation, while inspired by these (explicitly citing WiscKey and LavaStore), adapts the concept with columnar blob formats, rewrite compactions, and MVCC-specific handling to suit CockroachDB's distributed transactional needs. link to the blog post: https://lnkd.in/gDzs5Hrb

46

New release (3.4.1) of the quarkus-micrometer-registry. It includes a new registry compatible with #Micrometer Prometheus Client V1.x. In a future Quarkusio 4.x version, this client will become available in Quarkus core. https://lnkd.in/d2gZEX-z

15

Just over a year ago, Flexpa released the first public Model Context Protocol server for FHIR. Today, Assistant Secretary for Technology Policy calls out MCP in the overview to HTI-5. A lot can change in a year.

35

2 Comments

If you're building a workflow engine or anything in agentic computing, being able to capture and serialize continuations makes life MUCH simpler - it means you can pause a program, stick it in storage, and resume it later, nbd. This is the simplest and most expressive approach, and also the most efficient. Nothing really comes close. Squint and you can see attempts to approximate this capability for special cases, for particular kinds of continuations. All well and good if you don't have access to this magic trick, but if you can just do the thing directly, that's even better!! For our workflows / agents engine in Unison, we just capture and store the continuation. I love it. The focus is on the engine itself, not "how do we half-approximate the rest of the program after this point where it needs to pause?" The implementation shakes out much simpler, and we can do it for all continuations, without caveats and without forcing the user to phrase the workflow in a stilted way. Just write regular straight-line code, with functions and function calls.

56

4 Comments

Good implementations of wide tilde marks have always been hard to find. And good luck if you want a wide tilde *under* an expression (which I sometimes do). Here's a comparison of wide tildes produced by various equation editors that highlights the differences in quality. Good interoperability is also hard to find, as made pretty clear by what happens when you try to simply copy/paste one of these expressions into another application like Adobe Illustrator.

18

Excited to present a session on the evolution of Elasticsearch's Query DSL and its future! Explore new alternatives like retrievers for hybrid search, and discover the capabilities of the emerging ES|QL, including JOINs and natural language integration. Learn how to build multi-stage retrieval and enhance results by integrating LLMs directly into your ES|QL queries. Join us to stay ahead in the search landscape: bit.ly/3VfJ3nt  

27

2 Comments