ShortSingh — News in Short

Zerohash Reapplies for OCC Trust Bank Charter With Narrower Scope

Crypto infrastructure firm Zerohash has reapplied to the Office of the Comptroller of the Currency (OCC) for a trust bank charter. This marks the company's second attempt after its initial application failed to receive approval. The firm stated that this new effort would be more narrowly focused than its previous bid. Zerohash is seeking the federal charter to strengthen its regulatory standing within the U.S. financial system.

0 comments Read more at CoinDesk

ProgrammingDEV Community ·

Engineers Propose Adaptive AI Disclosure System to Balance Transparency and Persona UX

A technical discussion on DEV Community outlines engineering approaches for maintaining persistent AI identity disclosure in chatbot personas without degrading the user experience. The article argues that both one-time disclaimers and per-message repetition are ineffective, as the former gets forgotten while the latter becomes noise users tune out. The proposed solution uses risk-weighted disclosure frequency, injecting reminders more often during emotionally sensitive or high-stakes exchanges and less during routine ones. Disclosure language is varied in phrasing and woven into the persona's voice to avoid feeling mechanical, while a persistent UI badge provides a structural, always-visible signal independent of conversation timing. The system also couples disclosure logic with escalation detection, ensuring crisis scenarios trigger mandatory, explicit AI identification alongside appropriate resources.

0 comments Read more at DEV Community

ProgrammingGitHub Blog ·

GitHub shares lessons on evaluating LLMs before deploying to production

GitHub has published a blog post outlining key lessons learned from evaluating large language models (LLMs) for real-world use cases. The insights stem from the company's hands-on experience applying LLMs to secret scanning tasks. The post aims to help developers and teams assess LLM performance before moving models into production environments. GitHub's guidance reflects practical challenges encountered during internal evaluation processes.

0 comments Read more at GitHub Blog

ProgrammingDEV Community ·

Developer Credits Nutrition Deficiencies for Slowing Burnout Recovery Twice

A software developer experienced two burnouts within three years, following standard recovery advice both times — including taking time off, setting boundaries, and addressing psychological factors. Recovery remained unusually slow until bloodwork revealed critically low vitamin D, low RBC magnesium, an omega-3 index of 3.1%, and elevated inflammation marker hs-CRP at 2.9 mg/L. The developer explains that low vitamin D impairs dopamine synthesis, while low magnesium disrupts the HPA axis stress-response system, both of which can hinder burnout recovery. After supplementing with D3+K2, magnesium glycinate, and EPA-focused omega-3s over roughly eight weeks, cognitive and motivational symptoms reportedly improved. The developer notes that neither burnout prompted any clinician to suggest checking these nutritional markers, and has reported no recurrence in over two years since addressing them.

0 comments Read more at DEV Community

IndiaTimes of India ·

Supreme Court shields Ram Temple donation scam whistleblower from arrest

The Supreme Court has granted protection from arrest to a whistleblower who first raised allegations of misappropriation of devotee donations at the Ayodhya Ram Temple. Senior advocate Pradeep Rai appeared before a bench led by Chief Justice Surya Kant on behalf of the petitioner, identified as Upadhyay. Rai argued that the Uttar Pradesh Police was acting vindictively against his client in connection with a road accident case and charges filed under the SC/ST Act. He urged the court to transfer the investigation into these cases to an independent agency, citing concerns over a fair probe. The court's intervention comes amid broader scrutiny over the handling of funds donated by devotees to the temple trust.

0 comments Read more at Times of India

ProgrammingDEV Community ·

How to Safely Compare AI Text Classification APIs Using Tenant-Based Migration

A software engineering approach outlines a structured method for comparing OpenAI, Claude, and Gemini text classification APIs without disrupting live CRM workflows. The design migrates one tenant cohort at a time, keeping only a single authoritative path permitted to write CRM actions while candidate models run in shadow mode returning validated JSON. Output is restricted to a closed vocabulary of predefined tags, and any malformed or unrecognized responses are rejected before comparison begins. Per-tenant cost tracking is maintained throughout by recording production and shadow usage separately under the same tenant ID with different workload classes. Regional data processing policies and model permissions are tied to cohort state rather than prompt configuration, making the migration auditable and testable at each stage.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Google Adds 'Select from Screen' to Gemini in Chrome for Targeted Page Analysis

Google has introduced a new feature called 'Select from Screen' for Gemini in Chrome, allowing desktop users to draw a box around a specific area of a webpage and send it directly to the AI assistant for analysis. The selected region can include text, images, or a combination of both, giving users more precise control over the context they provide to Gemini. Users access the feature through Chrome's side panel, where they choose the option and manually define the area they want to share. The update aims to reduce friction in AI-assisted browser workflows, benefiting use cases such as product research, creative review, and document analysis. Availability may vary depending on account tier, region, rollout phase, and administrator settings in managed environments.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

AI Agent Built to Shop, Audit, and Pay in USDC Without Human Approval

A developer built a fully autonomous AI procurement agent for Google's All Things Agentic Hackathon that can compare prices across data providers, review its own spending decisions, and execute real USDC payments on-chain. The system uses two AI models — a primary decision-making agent and a separate auditor — but relies on a deterministic budget guard that checks confirmed on-chain spend from the past 24 hours to enforce a hard spending cap. After each payment, the agent independently verifies the transaction by reading the blockchain's Transfer event directly, rather than trusting the payment SDK's reported hash. Payments are executed using the x402 HTTP payment protocol with EIP-3009 signed authorizations, requiring manual reconstruction of low-level cryptographic payload details not exposed by standard libraries. Because the only real x402 provider tested required Base mainnet USDC, the team built their own testnet-compatible provider to validate the full payment loop without spending real funds.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

How Workflow, Auth, Taxonomy, and Memory Services Complete an AI Eval Platform

A developer building an AI evaluation platform details how four often-overlooked backend services — workflow orchestration, a taxonomy engine, a platform authentication layer, and evaluation memory suites — form the critical connective tissue that makes the system functional. The workflow orchestrator manages multi-step evaluation pipelines using a directed acyclic graph model, pausing at human review gates where judgment calls are required before the process continues. An early polling-based status system was replaced with WebSocket connections backed by Redis Pub/Sub, enabling real-time updates for concurrent reviewers across multiple cloud instances. Unstructured print-based logging was also overhauled with structured JSON logging that includes correlation IDs, allowing engineers to trace failures across services with a single query. Together, these infrastructure decisions transformed a loose collection of microservices into a coherent, auditable, and production-ready platform.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Dev builds Gradle plugin to cut KMP iOS export bloat, trims 5.7 MB in minutes

A developer optimizing a personal Kotlin Multiplatform app discovered that 61% of its 459 Objective-C exports had no Swift call sites, generating thousands of lines of unnecessary bridging code. To investigate, he built a Gradle plugin called kmprofiler that parses the generated Objective-C header and flags declarations never referenced in Swift source files. The bloat stemmed mainly from public theme constants, internal UI state models, and third-party library types leaking into the header via exposed function signatures. By marking just five files as internal in the shared Kotlin module, he reduced the Shared.h header from 23,622 to 13,327 lines and cut the compiled framework object size by 5.7 MB. The plugin, kmprofiler v0.1.0, has been published to the Gradle Plugin Portal for other KMP developers to use.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Dev Team Builds Deterministic Product-Swap Tool for Ad Editing, Limits AI Use

A software development team at Variant Multiplier tackled a real production request to replace one hearing-aid product SKU with another across an entire finished advertisement. Rather than letting an AI model rewrite the full ad — which risked altering untouched elements like music, voice quality, and pacing — engineers designed a deterministic planning system that only calls the model when a rule-based substitution cannot resolve a line. The work was delivered across five pull requests, starting with UI fixes to the transcript editor that allowed sentence-level selection and full-line retyping. A product catalog was then introduced to give the tool a formal concept of product identity, synced from a central source to avoid data duplication. The final approach produces a per-line swap plan that editors can review before any re-render is triggered, keeping AI intervention narrow and deliberate.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Developer Cuts AI Voice Pipeline Costs 7x by Fixing Double-Billing and Hardcoded Pricing

A developer audited and rebuilt an AI video platform's voice and lip-sync pipeline across ten pull requests, uncovering systemic billing flaws that caused costs to diverge from actual vendor invoices. Key issues included hardcoded per-call constants that ignored real token counts returned by Anthropic, and duplicate billing on retries where the vendor returned a cached asset without performing new synthesis. A faulty cache system also served outdated voice recordings when settings like emotion or TTS model changed, because the cache key only tracked scene index rather than content. Fallback chains for lip-sync SKUs were restructured to degrade upward to better-quality tiers on failure, preventing silent quality drops or accidental use of the most expensive options. Together, the fixes achieved a roughly sevenfold cost reduction on lip-sync processing without any change to output quality.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

AI Video Ad Platform Uses Vision-in-the-Loop to Auto-Fix Scene Prompts

A developer at an AI video ad platform has built a system called vision-in-the-loop that automatically reviews and rewrites prompts based on the actual frames generated by an AI video model. Previously, operators had to manually inspect each scene, identify visual errors, and rewrite prompts by hand — a process consuming up to an hour of attention per ad creative across multiple scenes and iterations. The new system captures the first keyframe of each generated scene and passes it to a vision model alongside the original prompt, scene intent, and brand constraints. The model returns a structured critique flagging specific issues such as incorrect framing, missing products, color palette drift, or avatar identity mismatches. A rewrite stage then applies targeted prompt corrections and triggers a single automated regeneration before the operator ever opens the review portal.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

How One Dev Fixed AI-Generated Ads to Match the Feel of Viral Originals

A developer at a video generation platform identified eight key failure modes causing AI-produced ads to feel visually inconsistent with their viral reference clips. Problems included avatars changing appearance between scenes, blank caption boxes appearing from unrecognized overlays, and missing ambient audio that gave originals their immersive quality. The fixes involved propagating a single avatar identifier across all scenes, rewriting caption rendering to display real styled text and skip undescribable overlays entirely, and bundling a dedicated font for consistent typography in production containers. These changes were implemented across roughly 2,800 lines of code spread over several pull requests. The result was a pipeline that moved from structural correctness toward perceptual fidelity, making generated ads feel like they belonged to the same family as the originals.

0 comments Read more at DEV Community

IndiaTimes of India ·

Supreme Court rules all Banke Bihari temple donations belong to deity

The Supreme Court has ruled that all donations made at the Banke Bihari temple must go to the deity. Senior advocate Maninder Singh informed the court that sevayats have been obstructing legitimate donation collection by covering the temple management committee's donation box with flowers. This alleged practice diverts devotee contributions away from the official box and into the hands of the sevayats. The case highlights an ongoing dispute between the temple's sevayats and its management committee over control of donations.

0 comments Read more at Times of India

ScienceWIRED ·

Labor Day Sales Bring Discounts on Instant Pots, Cookware, and Coffee Gear

Early Labor Day deals are now available on a range of kitchen products, including Instant Pots, cookware, and coffee equipment. The discounts come as families prepare for the return of school and sports seasons in fall 2026. Retailers are offering these promotions ahead of the official Labor Day holiday. The sales are aimed at helping households stock up on kitchen essentials during a busy seasonal transition.

0 comments Read more at WIRED

ProgrammingDEV Community ·

How a Discovery Framework Stopped a Startup From Building a $25K Unvalidated MVP

A startup founder with 26 years of enterprise experience approached a developer with an eight-phase, $23–27K fixed-price proposal to build an AI-powered organizational intelligence platform. The founder had already purchased a Dell R740 server with 512GB RAM for on-premises infrastructure before securing a single client or validating market demand. The developer declined to begin building and instead conducted a technical discovery session, which revealed the core product was essentially a document versioning system — a problem already addressed by existing tools like SharePoint, Confluence, and Git. Discovery also exposed that the on-premises infrastructure rationale, compliance requirements for defense and telecom clients, had no signed clients or pilot agreements to justify it. As a result, the engagement shifted from a build contract to an advisory relationship, with the founder agreeing to deprioritize the server purchase and redirect budget toward customer validation before writing any application code.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Developer Ships 8 UX Fixes in 48 Hours to Reduce Admin Dashboard Friction

A developer completed eight pull requests over a single weekend to improve the usability of an admin dashboard for a task seeding platform. None of the changes addressed broken functionality; instead, they targeted accumulated friction such as cluttered toolbars, hardcoded filter lists, and a search field that triggered on every keystroke. A key improvement replaced scattered, hardcoded status filter options with a single centralized registry, making it easier to add or remove filters without touching multiple files. Search behavior was also refined by setting a minimum query length of three characters, preventing noisy results from short inputs. The sprint demonstrated that meaningful UX gains can come from a series of small, deliberate changes rather than a full redesign.

0 comments Read more at DEV Community

WorldBBC World ·

Trump Claims Strait of Hormuz Fully Cleared of Mines

US President Donald Trump has announced that all mines have been cleared from the Strait of Hormuz. Iran had closed the critical waterway following the outbreak of conflict in the region. The closure of this vital shipping lane had triggered significant volatility in global oil prices. The Strait of Hormuz is one of the world's most strategically important maritime passages for oil transit.

0 comments Read more at BBC World

ProgrammingHacker News ·

XCancel shuts down after receiving cease-and-desist letter from X Corp.

XCancel, a third-party Twitter/X proxy service, has suspended its operations following a cease-and-desist letter from X Corp. The letter was received on Monday, August 24th at 8PM EST. The service's operators announced the shutdown on the XCancel website, stating it would remain offline until further notice. The team said they are currently seeking legal advice and are unable to share additional details at this time. XCancel had been operating for approximately two years before the shutdown.

0 comments Read more at Hacker News

← NewerPage 347 of 3473Older →