Feynman Liang

Is Elixir observable enough for production systems at scale? When senior and staff engineers evaluate a new runtime, the real question isn’t syntax or developer happiness, it’s operability. Can you inspect a live system under pressure? Can you correlate logs, metrics, and traces? Can you reason about failures without guesswork? I’ve published a deep-dive on the current state of observability in Elixir, aimed at engineers coming from JVM, Go, or Node.js backgrounds. It covers BEAM runtime introspection, structured logging, OpenTelemetry, Prometheus/Grafana, SaaS integrations, and, just as importantly, the trade-offs. No hype. No tutorials. Just an honest look at how Elixir actually behaves in production. 👉 Is Elixir’s Observability Ready for Production? https://lnkd.in/eSPeDNkD #Elixir #MyElixirStatus #Observability

21

Measuring developer productivity is complex. At Meta, the team has been taking a rigorous, data-driven approach and even building new infrastructure to support it. Kudos to Payam, Pavel, and the entire team for this impressive work and presentation. Over the past year, my main focus has been on building AI coding agents, so this really resonates. Check out the section about Devmate at 6:52 in the video to learn more on what I've been up to. A few highlights: - Thanks to Devmate, Meta’s developers are more productive - with Diffs per Developer per Month (DDM) improving by 6–12% (a conservative estimate!). - Devmate tracks AI code attribution at the character level, allowing Meta to know precisely which code was generated by Devmate vs. written by humans.

15

I've been discussing this recently with Frances Buontempo in the context of coding assistants. We hear a lot from developers saying that LLM assistants in their fave IDEs are great for generating a lot of the (boring) boilerplate code needed to stand up a non-trivial system. However, most popular IDEs have been able to do exactly this for decades--whether with "wizards", code-block and project templates, or other various forms of meta-programming. The difference is that the "old" way is much less hungry of resources, while simultaneously being much more predictable and reliable at automatically generating the "boring" stuff. Coding assistants are also good at other things, of course, but many of the popular (in my conversations with developers) tasks are those I would much prefer to do myself, such as writing tests.

15

6 Comments

Why I built my own SOTA file and Git based RAG and memory system for LLMs? (mostly just because it was fun) I wanted my local AI tools to have persistent memory, but I prefer to own my data. Most RAG solutions lock your data into opaque vector databases or require massive dependencies like PyTorch just to parse a PDF. I just want to drop all my documents in a folder for RAG. If my AI agent "remembers" a fact about me, I want to see it in a file. If it hallucinates, I want to edit that file. So I built `agent-cli memory` and `agent-cli rag`. It’s a local RAG and Memory proxy that sits between your chat interface (Cursor, Open WebUI, Cline) and your model. Why do I claim state-of-the art (SOTA)? Because uses well established techniques (e.g., two-stage ranking with cross-encoders) but re-implements them using ONNX Runtime to keep it lightweight. The key features: • File-based: Drop documents into a folder; they are instantly indexed. • Git-Versioned Memory: Every fact the AI learns is a simple Markdown file committed to a local git repo. • Universal: Works with any OpenAI-compatible client. I wrote a detailed post about the architecture, why I chose ONNX over PyTorch, and how the memory reconciliation loop works. Read the post: https://lnkd.in/gqimzxY5 Code: https://lnkd.in/gBFNi34S #Python #LocalAI #RAG #OpenSource #AI

40