SShortSingh.
TopTechnologyAI & MLStartupsProgrammingScienceBusinessFinanceCrypto & Web3World

Programming

LatestTopDiscussed
0
ProgrammingDEV Community ·

Developer open-sources CerebrumKit, moving AI agent configs from code to database

A developer has released CerebrumKit, an open-source framework that stores AI agent topology — including tools, skills, prompts, and workflow graphs — in a database rather than in source code. The shift means that non-technical domain experts can edit agent instructions and tool descriptions directly through a panel without requiring a new deployment. Each component, from individual tool specifications to multi-agent workflow routing, is stored as database rows and takes effect on the next message after editing. The project deliberately avoids chat-history replay in prompts, relying instead on keyed memory and well-designed tools to keep runs cheaper and more predictable. The author acknowledges trade-offs, including unsandboxed tool execution and limited support for complex programmatic control flow, recommending the tool only for appropriate use cases.

0 comments Read more at DEV Community
0
ProgrammingHacker News ·

Swedish startup builds AI strike drone that selects and attacks targets autonomously

A Swedish startup has developed an autonomous attack drone capable of independently identifying and striking targets without any human input. The system is powered by Nvidia's Jetson Orin Nano, a compact AI processing module, running a small on-device machine learning model. Notably, the drone requires no external communications, making it resistant to jamming or signal interference. The development raises significant ethical and regulatory questions about the use of fully autonomous lethal systems in modern warfare.

0 comments Read more at Hacker News
0
ProgrammingDEV Community ·

AI Era Raises Software Supply-Chain Risks as Untrusted Code Enters Builds Faster

The rise of AI coding assistants and autonomous agents has significantly accelerated the rate at which unvetted packages, models, and dependencies enter software build pipelines. Risks include AI tools hallucinating non-existent package names that attackers can register, third-party pip install instructions bypassing org policy, and coding agents silently installing packages at scale. Experts argue that effective supply-chain management boils down to four operational controls: provenance of resolution, identity management, publish authority, and active enforcement at download time. Visibility tools like SBOM generation and CVE reports are insufficient without serve-time policy enforcement at the registry level. Engineering leads are advised to route all package resolution through a single policy-controlled proxy per ecosystem, prioritize internal namespaces, and block malicious findings at artifact fetch rather than relying on dashboard alerts.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

Incident Context Agent Traces Production Failures Using Connected Knowledge Graphs

A developer has built 'Incident Context,' an incident-investigation agent submitted to the Sanity Challenge, designed to diagnose production outages without relying on guesswork or keyword matching. The tool models operational records — including services, deployments, configuration changes, and runbooks — as interconnected Sanity documents forming a structured graph. When investigating an outage, the agent traverses relationship paths between these documents to build a verifiable evidence trail rather than inferring causation from surface-level correlations. In a demonstration comparing two production incidents, the agent correctly identified distinct root causes: one tied to a database pool size reduction and another to a payment timeout configuration change. Every investigation output includes confirmed evidence, any inferences made, source paths for verification, and a recommended next step, with the system refusing to generate unsupported answers if its knowledge base is unreachable.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

Developer fixes vLLM KV-cache offloading bug affecting DeepSeek and Gemma-4 models

A software engineer identified and fixed a bug in vLLM, the open-source LLM inference engine, related to KV-cache offloading for models with mixed KV-cache groups. The existing logic assumed a single KV-cache group layout and used a single block_size value for chunking calculations, which broke for architectures like DeepSeek-V4-Flash and Gemma-4. The fix introduced a new parameter, blocks_per_chunk, to handle varied chunking requirements while preserving backward compatibility for existing users. The patch was submitted as PR #48878 to the upstream vLLM repository on GitHub. The case highlights how LLM infrastructure bugs can stem from outdated internal assumptions rather than visible runtime failures.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

Developer Runs AI-Automated Blog for 19 Days Straight, Approves Pipeline Once

A developer at mrsaynothing.dev built an AI-powered pipeline that has autonomously written, translated, and deployed a blog post every day for 19 consecutive days across 18 languages. The system runs on three scheduled jobs on a single desktop machine, handling writing, deployment, cross-posting, and translation without daily human input. In its first seven days of tracked analytics, the site recorded 146 visitors and 288 pageviews, with one non-tutorial post on RAM outperforming the tutorial median by more than five times. The developer logged four failures over the 19-day period, attributing all of them to flawed human specifications rather than model errors. The experiment highlights that managing an agent-driven workflow is less about building impressive AI and more about writing precise, unambiguous rules that leave no room for misinterpretation.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

Developer Builds Rate-Limiting System After One User Drained Telegram Bot's Daily AI Quota

A developer running a Telegram bot powered by Google's Gemini free-tier API faced a critical outage after a single user sent 1,400 messages in three hours, exhausting the 1,500 daily request limit and leaving over 999 other users unable to access AI features. To prevent recurrence, the developer implemented a three-layer protection system consisting of per-user rate limiting, a global daily request cap, and an in-memory response cache for frequently asked questions. The caching layer alone reduced quota consumption by an estimated 50–70%, with no additional infrastructure or database costs. Following these changes, 99% of users were reported as unaffected during subsequent bot usage. The experience highlighted a broader vulnerability in free-tier AI APIs, where no built-in safeguards exist against individual users monopolising shared daily quotas.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

Practical Guide to Starting a Blog and Growing an Audience From Zero

Building a blog audience requires a focused topic, consistent publishing, and a basic reader-acquisition strategy, according to a practical guide published on DEV Community. Writers are advised to choose subjects they genuinely understand or enjoy rather than chasing popularity, and to target a specific niche audience instead of writing broadly. Starting with a goal of publishing 10 useful articles helps bloggers learn what resonates before worrying about scale or platform choice. Readability matters as much as content quality, so short paragraphs, clear headings, and real examples are recommended to keep readers engaged. Basic SEO practices — such as descriptive titles and search-intent awareness — can improve discoverability, but the guide emphasises writing for people first and search engines second.

0 comments Read more at DEV Community
0
ProgrammingHacker News ·

Radius relaunches as open community events platform to rival Meetup.com

A developer has officially relaunched Radius, a community events platform built as an alternative to Meetup.com, after an extended development period following an initial debut on Hacker News. The platform focuses on helping people discover local communities, groups, and events, incorporating feedback gathered from over 150 comments during its first showing. A new feature called Activities has been introduced, allowing individuals to post lightweight, group-independent events — such as a casual bike ride — where others can join and vote on a suitable time. Radius is built using Ruby on Rails and is currently live at radius.to. The developer has indicated future plans may include open sourcing the project and exploring federation capabilities.

0 comments Read more at Hacker News
0
ProgrammingHacker News ·

Historical Study Documents Chiming Machines at Gloucester Cathedral and City Churches

A historical paper published by the Bristol and Gloucestershire Archaeological Society examines the chiming machines installed at Gloucester Cathedral and its surrounding city churches. The study, authored by MacKechnie-Jarvis, is part of volume 135 of the society's transactions. The document traces the mechanical history of these timekeeping and bell-ringing instruments across the ecclesiastical buildings of Gloucester. It offers a detailed account of how these devices were used and evolved within the city's religious institutions.

0 comments Read more at Hacker News
0
ProgrammingDEV Community ·

Codename One Fixes Severe HashMap Miss Performance Bug With Perturbed Probing

Developers working on Codename One, an open-source cross-platform Java/Kotlin framework, discovered a critical performance flaw in ParparVM's HashMap where lookups for absent keys were dramatically slower than for present ones. Benchmarks revealed that three million containsKey calls for missing keys took 32.7 seconds, while probes per miss reached over 222,000 for tables with one million entries. The root cause was that integer keys clustered in adjacent slots, causing linear probing to walk long uninterrupted runs before finding an empty slot. The team resolved this by adopting a CPython-inspired perturbed probing recurrence, reducing mean probes per miss to under two across all tested table sizes. Additional improvements included using cached hash codes to short-circuit unequal string comparisons and routing compatible UTF-16 string comparisons through native memcmp for faster vector-level processing.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

How Docker Sandboxes Redefine Security Boundaries for AI Coding Agents

Docker Sandboxes introduce a structured isolation model for AI coding agents by running them inside microVMs with their own kernel and Docker Engine, limiting direct host access. The model offers three file-sharing modes — direct mount, clone, and mountless — each carrying different levels of risk depending on how much the agent can interact with the host repository. Credential exposure is reduced by routing provider keys through a host-side proxy, though the agent retains whatever authority those credentials grant. Network traffic defaults to a deny-by-allow policy, but preset allowlists like the 'Balanced' profile may still permit broad access that exceeds task requirements. Shared skill stores and MCP gateway configurations can also extend trust boundaries across sandboxes, making it critical to audit each capability boundary before deployment.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

Developer Completes Google Cloud GenAI Academy APAC Cohort 3, Earns Gen AI Master Certificate

A developer completed the Google Cloud GenAI Academy APAC Edition Cohort 3, a program organized by Hack2Skill in collaboration with Google Cloud, earning a Gen AI Master Certificate upon completion. The program covered three tracks focusing on building customer-facing AI agents, turning business data into strategic decisions, and automating daily operations using productivity agents. Participants gained hands-on experience through workshops, codelabs, and quizzes, culminating in an Ideathon Prototype Submission requiring a working project. Key technologies explored included Google's Agent Development Kit, Gemini, Retrieval-Augmented Generation, BigQuery, Model Context Protocol, and Cloud Run. The program offered the developer their first structured exposure to how AI agents can integrate with cloud infrastructure, external data sources, and real-world applications beyond simple chatbot functionality.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

Skill Recorder keeps all data local until Analyze sends it to GitHub Copilot

Skill Recorder is a tool that captures screen activity, window titles, browser URLs, clipboard previews, and optional voice narration to help users automate tasks via GitHub Copilot. All recording, transcription, and frame extraction happen locally on the user's device, and no data leaves the machine until the user clicks the Analyze button. At that point, extracted screen images, the event timeline, and narration text are uploaded to GitHub's cloud for Copilot to process and convert into reusable Skills or Automations. The app's README explicitly warns users never to record, type, paste, or narrate passwords, API keys, or sensitive credentials, though it does not confirm whether such secrets are automatically redacted. Key details — including local storage location, data retention periods on GitHub's servers, and how to delete saved sessions — are not addressed in the README, raising unresolved questions for anyone considering use on sensitive or customer-facing systems.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

Developer Builds Open-Source Agile Tool Agila After Rejecting OpenProject and Vikunja

A founder managing a small, private secure-communications product in 2023 grew frustrated with existing project management tools like OpenProject and Vikunja, finding them either too heavy or a poor fit for daily workflows. Jira was ruled out early as overly complex for a lean team of four. Using AI-assisted coding, the founder built a minimal Kanban tool tailored to the team's actual needs, which eventually evolved into Agila. Agila is now a free, open-source, multi-view agile project management platform supporting Kanban, list, Gantt, and calendar views, with features like real-time updates, optional sprints, SSO, and soft WIP limits. It can be self-hosted or used via a hosted plan, and the full codebase is publicly available on GitHub.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

How to Build a Live Sports Scoreboard Using Vanilla JS and WebSockets

A developer tutorial published on DEV Community walks through building a real-time sports scoreboard using the Orbistats API, plain JavaScript, and WebSockets — no frontend framework required. The guide covers fetching today's fixtures via a REST endpoint, rendering match cards with live score displays, and opening a persistent WebSocket connection for real-time score updates. A key focus is handling socket disconnections gracefully, with an automatic three-second retry loop to prevent scoreboards from silently going stale. The project uses Orbistats' free-tier API key, which allows up to 150 requests per day across all endpoints with no credit card needed, and a public sandbox is available for testing without signup. The finished component is designed as a drop-in module that can later be adapted into React or Vue applications.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

React Derived State: Storing Computed Values in useState Causes Silent Bugs

A common React coding pattern — storing a filtered or computed list as a separate state variable and syncing it via useEffect — introduces subtle bugs where the UI briefly shows stale data. This happens because Effects run after the DOM commits, causing an extra render cycle before the correct value appears on screen. The root issue is treating a value that can be calculated from existing state or props as if it were independent state that needs to be stored. React developers are advised to compute such derived values directly during render, reserving useMemo only for genuinely expensive calculations. The article, written against React 19.3 and the stable React Compiler 1.0, outlines how to identify and eliminate this pattern across common and edge-case scenarios.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

Why Browser Drag-and-Drop Silently Fails and How to Fix It

A common but non-obvious bug in HTML drag-and-drop causes dropped elements to snap back with no console error, leaving developers puzzled. The root cause is that browsers block drops on all elements by default, and the only way to unlock a valid drop target is to call e.preventDefault() inside a dragover event listener. Without this single line, the drop event never fires regardless of how correctly the rest of the code is written. A secondary issue involves flickering highlight states on drop zones that contain child elements, caused by dragenter and dragleave events firing at every DOM boundary crossing. This can be resolved by checking the leave event's relatedTarget to confirm the cursor has actually exited the drop zone entirely.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

What Causes Nuxt Hydration Mismatches and How to Fix Them

Hydration mismatches in Nuxt occur when the HTML rendered on the server differs from what Vue produces on the client during JavaScript initialisation. Because Nuxt runs the same component twice — once in Node on the server and once in the browser — any value that depends on the environment, such as random outputs, dates, or localStorage, can trigger a mismatch. The bug can range from a barely noticeable flicker to broken interactivity, as Vue may discard and rebuild DOM nodes that users have already interacted with. Developers can resolve most cases using onMounted to defer client-only values, the ClientOnly wrapper component, or the data-allow-mismatch attribute for intentional differences. The guidance is written against Nuxt 4.x but also applies to Nuxt 3 running in compatibilityVersion 4 mode.

0 comments Read more at DEV Community
← NewerPage 1090 of 3217Older →