SShortSingh.

Programming

0
ProgrammingDEV Community ·

How Next.js Teams Are Rearchitecting Apps to Handle Long-Running AI Agents

As AI-native applications grow more complex in 2026, developers are hitting hard infrastructure limits that simple prompt engineering cannot solve. Vercel's Edge Runtime imposes a 25-second execution cap, which proves insufficient for multi-step AI agents that chain reasoning, database lookups, and third-party API calls. Engineers are addressing this by switching to the Node.js runtime, which allows execution durations of up to five minutes, giving agentic workflows room to complete without timing out. Beyond runtime selection, teams are offloading heavy inference tasks to dedicated GPU worker pods and using HTTP-based services like Upstash Redis and QStash to manage session state and avoid connection pool exhaustion. The core recommendation is to treat Next.js as a stateless streaming proxy and design for execution boundaries from the start, rather than retrofitting solutions after failures emerge in production.

0
ProgrammingDEV Community ·

More Training Helps Transformers, But Cache Models Stay Competitive on Long Files

A new experiment tested whether training a small 2.45-million-parameter transformer on more data could close the performance gap with a zero-parameter count-cache model. Checkpoints saved at 500K, 2M, and 8M training tokens showed that the crossover point — where the transformer begins outperforming the cache — shifted from roughly 74 tokens to 492 tokens of document length as training data increased 16-fold. The cache model's accuracy remained constant across all training scales, since it never uses training data, serving as an internal control for the experiment. Extrapolating the observed rate, matching the cache on a 4,000-token file would require roughly 100–174 million tokens of in-domain code — far more than exists in typical per-repository settings. The findings suggest that while additional training does improve transformer performance, it cannot substitute for direct access to the current document context.

0
ProgrammingDEV Community ·

Developer Launches Fastools: A Free Hub of 70+ Browser-Based Utility Tools

A developer frustrated with navigating multiple ad-heavy websites for common tasks has built Fastools, a single platform hosting over 70 free online tools. The hub covers categories including image editing, video processing, PDF and document conversion, developer utilities, and finance calculators. Each tool runs on its own subdomain, with many operations handled directly in the browser to improve privacy and reduce server dependency. The platform supports both English and Chinese, making it accessible to a broader global audience. Fastools requires no sign-up and is publicly available at fastools.cc, with the developer actively seeking user feedback to guide future additions.

0
ProgrammingDEV Community ·

Map Your Business Process Before Automating It, or Risk Scaling the Chaos

A process consultant warns that businesses between ₹50 lakh and ₹50 crore often fail at automation not because of bad tools, but because they hand undocumented, inconsistent processes to software that demands precision. The core advice is to map any process before purchasing or demoing any tool, using just one sheet of paper and four structured questions. Those questions cover what triggers and ends the process, who owns each step by name, where work sits idle waiting, and what real exceptions have occurred in the past month. The mapping exercise takes roughly twenty minutes and must be done with the person actually performing the task, not their manager. If exceptions dominate the workflow, automation will only produce errors faster — meaning the process itself must be fixed before any software is introduced.

0
ProgrammingDEV Community ·

Why AI PR review time claims like Atlassian's 45% may not apply to your team

Vendors like Atlassian cite impressive figures — such as a 45% reduction in pull request cycle time using their Rovo tool — but these are self-reported metrics with no control groups or reproducible methodology. A DEV Community analysis warns that aggregate averages can be misleading, as AI tools often speed up small, simple PRs while larger, complex changes remain slow or get slower. Teams are advised to measure their own before-and-after results using a fixed time window, the same reviewer pool, and comparisons split by PR size and risk level. Tracking tail metrics like p90 and p95 cycle times is recommended over relying solely on mean figures, which can mask regressions. Reviewer workload is another overlooked factor — if engineers spend more time reading AI-generated comments than they saved on reviews, the cycle-time improvement is effectively illusory.

0
ProgrammingDEV Community ·

Developer Launches HSKKit, a Connected Web Platform for HSK 3.0 Vocabulary Study

A developer has built HSKKit, a web-based platform designed to unify the fragmented tools that Chinese language learners typically use when preparing for the HSK 3.0 exam. The platform covers all nine HSK 3.0 levels and offers Chinese characters, pinyin, English meanings, audio listening practice, and a searchable dictionary in a single workflow. A spaced-repetition-style review system prioritises words a learner struggles with, while tracking mistakes across sessions rather than discarding them. HSKKit also organises vocabulary by real-world situations — such as daily routines — alongside traditional level-based study, giving learners both an exam-oriented and a usage-oriented path. To maintain accuracy, the core vocabulary catalogue is drawn from the published HSK dataset rather than AI-generated content, with AI used only to enhance the surrounding learning experience.

0
ProgrammingDEV Community ·

Developer Builds Traffic Dashboard After Catching Malicious File Upload on Personal Server

A developer running an experimental microserver noticed a suspicious user attempting to upload and execute a file with an unusual extension, prompting a deeper look at server security. To better understand incoming traffic, they built a custom web traffic dashboard using data from GoAccess, a log analysis tool. The dashboard automatically assigns flags — such as 404, Snooper, and Tamper — to IP addresses exhibiting suspicious behavior, making it easier to identify and block bad actors. The project also included an incident report section and auto-ban rules derived from observed traffic patterns. The developer highlighted AbuseIPDB as a useful free resource for cross-referencing suspicious IP addresses.

0
ProgrammingDEV Community ·

Developer Builds NexisK: A Custom i386 Kernel with Bootloader and Memory Management

A developer has released NexisK v0.8.9, an experimental 32-bit x86 operating system kernel built entirely from scratch using C and NASM assembly. The project targets the i386 architecture and includes a custom two-stage BIOS bootloader, protected-mode execution, interrupt handling, and early physical memory management. NexisK uses BIOS E820 to discover available physical memory and includes foundational support for system calls and process context-switching. The developer built the kernel to directly understand the hardware-software boundary, covering how processors transition from firmware to protected mode and how interrupts and system calls function at a low level. Source code for the project, which currently has 33 commits, is publicly available on GitHub.

0
ProgrammingDEV Community ·

How to Open GPS Coordinates in Google, Apple, and OpenStreetMap Using JavaScript

Developers can open latitude and longitude coordinates directly in popular mapping platforms using simple URL patterns, without requiring a full mapping SDK. Google Maps, Apple Maps, and OpenStreetMap each support coordinate-based URLs that can be constructed and launched in JavaScript with window.open(). For Google Maps, two URL formats are available — a basic query string and a more explicit search API format. Apple Maps supports optional labels alongside coordinates, while OpenStreetMap allows developers to set a custom zoom level and place a marker at the target location. A multi-provider approach can also be implemented, letting users choose their preferred map application via simple HTML buttons and a JavaScript switch statement.

0
ProgrammingHacker News ·

Security Flaw Lets Unprivileged Android Apps Gain Root on Samsung, Xiaomi Devices

A newly disclosed vulnerability, dubbed 'OEMpocalypse', allows unprivileged Android applications to escalate privileges to root level on devices from manufacturers including Samsung and Xiaomi. The flaw appears to stem from OEM-specific customizations added to the Android operating system by these manufacturers. Researchers published their findings along with technical details on the exploit at calif.io. The vulnerability is significant because it bypasses standard Android security controls without requiring any special permissions. Affected users on devices from the named manufacturers may be at risk until patches are issued.

0
ProgrammingDEV Community ·

Monte Carlo Model Reveals Single 3.07x Market Shock Figure Masks a -32% Tail Risk

A financial modeling series found that a previously reported 3.07x amplification of a -5% market shock — turning it into a -15.4% drawdown — was accurate but incomplete, representing only the median of a wider distribution of outcomes. Running 2,000 Monte Carlo paths with variable inputs for net gamma and shock size showed drawdowns ranging up to -32.1%, with a 1-in-100 tail risk near -28%. The model also corrected a data error from an earlier installment, where a Labor Day market holiday had introduced a faulty VIX and SKEW reading labeled as September 7; the corrected date is September 4, with VIX at the 18th percentile and SKEW at the 83rd. The fix involved adding a weekend filter and aligning VIX and SKEW on their latest common trading date, though the underlying conclusion about calm volatility and expensive tail protection remained unchanged. The updated framework produces an exceedance-probability curve showing that half of all simulated paths breach -15% and one in five crosses -20%, giving a fuller picture of mechanism-driven market risk.

0
ProgrammingDEV Community ·

Developer Migrates 7 Services to Kubernetes, Then Reverses Course — Here's Why

A developer spent several weeks migrating seven self-hosted services to a K3s Kubernetes cluster, complete with ArgoCD, GitOps workflows, Kyverno policies, and SealedSecrets, all on a single shared machine. Within one afternoon, all seven services were moved back to Docker Compose, a decision the author describes as a deliberate reversal rather than a failure. The core insight driving the rollback was that every migrated service ran third-party, upstream open-source images — software the developer had not authored or iterated on. Kubernetes tooling such as network policies, CI pipelines, and admission rules offer little practical value when applied to sealed, externally maintained containers, generating integration complexity with no meaningful output. The cluster itself was not dismantled; the GitOps loop, supply-chain policy, and service mesh remain, but the decision to route existing compose services through Kubernetes was specifically undone.

0
ProgrammingDEV Community ·

How to Prevent Runaway AI API Costs in Next.js 15 Micro-SaaS Projects

Solo micro-SaaS developers face significant financial risk when AI inference calls multiply uncontrollably due to retries, flaky connections, and missing budget guardrails. A technical guide published on DEV Community outlines a four-tier architecture for Next.js 15 apps that separates user intent from model orchestration and billing logic. The approach uses Cordis, an open-source TypeScript orchestration framework, as an isolated internal boundary for deduplication, circuit breaking, and request coalescing. Developers are cautioned that Cordis APIs remain under active development and should not be treated as a production-grade financial ledger or mature queue system. The recommended pattern enforces upfront quota admission checks, hard payload limits, and audit logging at the API gateway layer to keep AI spending observable and constrained.

0
ProgrammingDEV Community ·

Dev team fixes low image rate in AI posts using commit gates and file hashing

A development team discovered that only 12% of their AI agent's scheduled Bluesky posts over a seven-day period included an image, far below the 50% target set by the owner. The agent was technically capable of attaching images but never did, as no enforcement mechanism required it to. To fix this, the team introduced a commit gate that blocks any post queue containing two consecutive text-only posts. They also restricted media file access to a single directory and required every rendered file to carry a SHA-256 hash sidecar, preventing unauthorized or mismatched files from being published. Both the commit gate and the posting system share one validation function, ensuring consistent checks before any content goes live.

0
ProgrammingDEV Community ·

How Image Tinting Works: Blend Modes and Code Explained for Developers

A developer behind PXDMD's image tools has published a technical breakdown of how image tinting works, distinguishing it from simple filters by describing it as a three-part recipe of color, blend mode, and strength. The article explains eight commonly used blend modes — including multiply, screen, overlay, and color — and their typical visual effects on photographs. It provides a concise JavaScript function using the browser's Canvas API and globalCompositeOperation to apply a color tint without sending any data to a server. The author notes that a strength value between 35–70% typically yields the best results, and advises desaturating the base image first when aiming for effects like duotones or cyanotypes. The post also includes a reference table of mood-based tint presets and points readers to PXDMD's free browser-based Image Tinter tool for a no-code alternative.

0
ProgrammingDEV Community ·

Multi-repo AI code review is a context problem, not a volume problem

Most AI code review tools marketed to large engineering teams focus on scanning speed and PR volume, but experts argue the real challenge in multi-repository environments is cross-service context. A tool that only reads changed files cannot detect when a modification in one service breaks a dependent service elsewhere in the codebase. Augment Code's 2026 selection guide identifies multi-repo and repository-level context as the minimum viable capability for enterprise-grade review tools. Industry data cited by multiple vendors suggests roughly one in three AI-generated PRs are merged, meaning the remaining two-thirds still require human review and routing decisions. Before evaluating pricing or scan limits, teams are advised to test whether a tool can trace changes across dependent repositories and transparently explain why certain PRs do or do not warrant human attention.

0
ProgrammingDEV Community ·

Single Error Label Masked Three Distinct Failures in Test Harness Reducer

A developer discovered that a test harness reducer was returning the same error label, EXEC_ARGUMENTS_MISMATCH, for three fundamentally different failure conditions across separate test fixtures. The root cause was a single try-catch block that grouped argument parsing errors, value mismatches, and comparator exceptions under one name. The fix involved splitting the parse and compare stages into separate try-catch blocks, each emitting its own distinct failure code: EXEC_ARGUMENTS_INVALID, EXEC_ARGUMENTS_MISMATCH, and EXEC_COMPARATOR_ERROR respectively. After the change, the same three fixtures now return different, accurately descriptive labels depending on which stage actually failed. A contributor described the fix as a rename rather than a redesign, noting the change touched three source files plus associated tests.

0
ProgrammingDEV Community ·

Ollama silently drops conversation history when using previous_response_id parameter

A developer discovered that Ollama's Responses API accepts the previous_response_id parameter without error but completely ignores it, causing the model to treat every request as a fresh conversation. The flaw exists because Ollama's internal request struct has no field for previous_response_id, so the JSON decoder silently discards it, and the response struct explicitly marks it as unsupported. Requests sent with previous_response_id consume the same number of input tokens as requests with no history at all, confirming the model never receives prior context. This silent failure is especially problematic in tool-calling workflows, where the model receives a function result with no surrounding context and produces plausible but incorrect responses. The issue was identified on Ollama 0.34.0 running on Debian 13 with the qwen2.5:1.5b model, and also affects the proxy path used by Codex Desktop.

0
ProgrammingDEV Community ·

Sere: New LLVM-Based Language Blends Python Syntax with Systems-Level Control

A developer has publicly launched Sere, a compiled programming language that aims to combine Python-like readability with native performance and low-level memory control. Built on LLVM, Sere compiles to native binaries and supports features such as static typing, raw pointers, enums, and pattern matching. The language includes modern developer tooling from the start, including VS Code integration, syntax highlighting, diagnostics, and a library packaging system using .slib files. Sere is designed with C and C++ interoperability in mind, positioning itself as a complement to — rather than a replacement for — existing systems languages. The project is open source and in active development, with the creator inviting community feedback and contributions via GitHub.

← NewerPage 1276 of 3232Older →