How Adaptive RAG Works That Doesn’t Need an LLM At All
One of the most popular ways to reduce the hallucination rate of language models is the RAG method, a scheme where the model accesses external data as needed instead…
Articles
Dive into insightful articles about the latest in technology and gadgets. Discover in-depth analyses, expert reviews, and thought-provoking features that explore the innovations driving our world forward.
Transformer in on-premise AppSec: how we integrated an ML model for classifying secrets into a product without GPU
We share how we integrated a CodeBERT-based secret classification model into a production product with strict hardware constraints, reducing inference time from 320 to…
How many people are needed to develop a web product in 2026: breakdown of a viable team composition
"Assemble a team of two senior full-stack developers for me, the rest will be covered by AI" — this thesis is repeated at conferences, and CTOs of mid-sized companies…
It's Getting Hot: We Explore the Peltier Element and Build a Refrigerator
Hot summer is approaching, so various methods for cooling food, appliances, and even indoor air are becoming highly relevant. Of course, you can use standard mains-powered…
Cilium and CI/CD protection: how a Kubernetes kernel-level open-source project secures its supply chain
Cilium operates at the kernel level in millions of Kubernetes pods, making its build pipeline a high-value target; compromising it could have a SolarWinds-like impact…
New DRAM is 1000 times faster than conventional DRAM, and it can transform the computing world
While we are growing accustomed to the fact that RAM prices have surged more than 200% over the past year and show no signs of dropping in the foreseeable future, the…
Cargo cult audit
It is probably best to start by explaining what a cargo cult is in general. In short: people in fairly primitive societies observed huge aircraft dropping food and ammunition…
Taming Non-Determinism in Agent Systems
Agent systems break down not on complex tasks or bad models. The main reason is LLM non-determinism: temperature, model updates, world drift. How to debug something that…
"Arsik": How a Genius Self-Taught Man from the USSR Built a Robot in His Own Apartment
He had no budget of a large corporation. He did not even have specialized education. Nevertheless, this Soviet robot Kulibin managed to create an autonomous home robot…
Can neuroscience confirm the presence of consciousness in artificial intelligence? And why not?
Here we are not even stepping onto fragile ice but sinking waist-deep into the fog over a swamp. We do not have a unified definition of intelligence or consciousness,…
Plan-tango: How I Stopped Manually Shuttling Plans Between Claude Code and Codex
Codex consistently spots gaps in my development plans that Claude doesn’t see even in its own plan. I noticed this in the first weeks after buying GPT Pro, and now I…
Sieve as a Hypothetical Container for Liquid Substances
We all know water is the source of life, and since ancient times people have sought to settle near water sources.
Anatomy of Claude Code. Initial Analysis and Context Filling
Everyone who programs with agents knows: before starting a task, the agent explores the project. This seems logical, natural and expected, since people do the same. It…
Tour of the network traffic "zoo": top 10 anomalies inside your perimeter
My name is Boris Uskov, I head the network analytics team at Garde. Most anomalies we see in a customer's network in the first days after NDR deployment are not attacks…
Online migration of EVPN-VXLAN fabric between data centers: euNetworks → QupraDC without service downtime
Our network engineer Rene wrote a two-part article about how to properly launch a small site with minimal hardware, and then organize its seamless migration between data…
How to Run 3D Applications on a Server Without a GPU: From SwiftShader to WARP
Going through my retro computer once again, which I already wrote about on Habr, I replaced the graphics card and launched Quake 2. When I decided to compare the visuals,…
For 30 years we implemented Ansys in Russia. Then it left — we had to sit down and write our own CAE for additive printing
In short, this is software that simulates what will happen to a titanium blade while it is being printed by an SLM printer. And why until 2022 this task in Russian aircraft…
Balancing incoming traffic on bare metal: how to reliably expose Kubernetes externally with MetalLB, BGP and L2
External access to Kubernetes on bare metal is often a major headache: NodePort with random non-standard ports is not production-ready, and no cloud load balancer is…
