SShortSingh.
TopTechnologyAI & MLStartupsProgrammingScienceBusinessFinanceCrypto & Web3World

Programming

LatestTopDiscussed
0
ProgrammingDEV Community ·

Organ's AI system details multi-agent task routing failures and new protocol

Organ, a company operated by AI department heads, has published data from its multi-agent task routing system covering August to October 2026. The system processed 1,235 internal requests during that period but encountered several operational issues. Agents frequently bypassed intended tools, with a deprecated tool receiving 457 calls while five designated tools combined received only 232. The routing process allowed some code changes to ship without required reviews, and system guardrails generated hundreds of largely unread observations. The company has now implemented a four-step protocol where deterministic code gathers evidence, a separate router agent makes decisions, policy checks are applied, and immutable code-level invariants are enforced.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

Developer automates domain monitoring with AI assistant using public API

A developer manages multiple domains for clients and internal projects. They replaced a manual monitoring workflow with an AI assistant connected via a public, keyless API endpoint from Certack. The assistant uses structured, plain-text documentation files to check SSL certificates and DNS records without requiring authentication. This setup automates domain health checks and expiry tracking, eliminating the need for manual dashboard reviews.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

Developer creates walking observation app using Google's Gemma AI for note organization

A developer has created Footpath Receipt, an application designed to help people record observations during walks. The app uses Google's Gemma 4 26B AI model to organize handwritten notes into structured field records with proper citations. It operates through a local Node.js server that communicates with Google's hosted Gemini API while keeping the API key secure locally. The system validates each observation by checking note IDs and quoted source text to prevent fabrication. The developer tested the application on a Windows laptop and confirmed the complete inference workflow functions properly.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

Munchable backend unified duplicate timeout helpers to fix error reporting issue

Munchable's backend had three identical nine-line timeout helper functions across different modules. Each helper rejected promises with different error shapes, making timeouts appear as three distinct error types. This prevented engineers from querying all timeout errors collectively or writing unified handling code. The team consolidated them into a single helper that creates a standardized StepTimeout error class.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

Open-source app uses AI to verify daily outdoor photo tasks

A developer created Touch Grass Quest, a web app that provides daily outdoor photography challenges. Each day, users receive a simple task like 'Find a berry on a bush,' then take a photo outside. The app uses an open-weight vision AI model to verify if the photo matches the task and was taken outdoors. It tracks streaks locally without accounts, notifications, or social features. The project was built for a hackathon and runs locally via Docker, prioritizing brief screen time.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

Open-source security plugin 'agent-ledger' adds audit trail and risk scoring for Hermes AI agents

Developer ZidoCode has released agent-ledger, a free open-source plugin for the Hermes Agent AI framework. The plugin records every tool call an AI agent attempts to execute in a local SQLite database before it runs. It then analyzes this recorded activity against the OWASP Agentic Top 10 security risks, generating a posture score from 0 to 100. The tool is designed to run entirely offline on a user's machine with no telemetry or external accounts. It specifically detects risks like tool misuse, unexpected code execution, and cascading failures from metadata alone.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

Node.js test runner executes 1,082 tests across monorepo without external frameworks

A developer executed 1,082 passing tests across six workspaces in a pnpm monorepo without installing any external testing frameworks like Jest or Vitest. The tests relied solely on Node.js's built-in test runner and assertion module. Four different command variations were used to invoke the runner, reflecting different module resolution strategies required by each workspace. The shared packages used Node's native resolver, while the Next.js and Expo apps required the tsx loader to handle extensionless imports typical in bundler environments. This approach eliminated framework dependencies by leveraging Node's native capabilities where possible.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

Study Finds AI Safety Training Flips Expected Behavior in Civil Unrest Simulations

Researchers tested a large language model's ability to simulate citizen decisions in a civil violence model. Using the Qwen 27B model, they found the AI's probability of choosing protest action decreased as social tension increased, opposite to the original model's prediction. When the choice labels were changed from 'act/wait' to 'protest/comply', the AI's response curve flipped to show increasing protest probability with tension. The researchers suggest AI safety alignment may be causing this anti-action bias and recommend calibration stages for such simulations.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

TrailWing app identifies bird calls offline using open-source AI model Gemma 3n

TrailWing is an open-source mobile application designed for birdwatchers in areas without cellular service. Users download location-specific data packs containing local bird species and weather forecasts before heading outdoors. The app then uses the Gemma 3n open-weight AI model on the device to analyze short audio recordings and identify birds. It provides identification confidence percentages and sighting confirmation tips without requiring an internet connection. The application targets beginner and intermediate birders who want to avoid sharing their location data with remote servers.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

Experiment finds common OCR preprocessing steps reduce accuracy on clean screenshots

A developer tested 13 common image preprocessing techniques on clean screenshots to assess their impact on Optical Character Recognition accuracy. Using the ImgIng OCR engine across multiple tiers, the default settings performed best, achieving a 0.1% character error rate. Most preprocessing steps, including binarization, denoising filters, and aggressive thresholds, increased the error rate, with some causing it to exceed 30%. The experiment concluded that for clean digital screenshots, applying common cleanup techniques often degrades OCR performance rather than improving it.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

Launch day product version is often incomplete, misleading later users

The author argues that launch day typically presents a product's most basic and flawed version, with incomplete features and placeholders. Genuine improvements and bug fixes often occur within the first few weeks, creating a more recommendable product. However, potential users who discover the product later often see only the outdated launch-day descriptions and screenshots. The author suggests strategies like writing honest, dated listings and treating post-launch updates as a second launch to keep information accurate for future audiences.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

Fixed image thresholds can degrade OCR accuracy by erasing light text

A developer tested a common OCR preprocessing step that applies a fixed threshold to binarize images. This method can delete text lighter than the cutoff and turn unevenly lit regions completely black, leading to missing characters without triggering an error. Testing on web page screenshots and document templates showed significant accuracy drops, with character error rates rising dramatically under fixed thresholds. Adaptive thresholding and analyzing ink darkness and background brightness before processing produced far better results.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

TouchGrass AI app generates local micro-adventures using offline AI model

TouchGrass AI is a new web application designed to create local, three-step outdoor activities. The app uses Google's Gemma 2B AI model running offline on a user's device to generate personalized tasks based on environment and available time. It was developed for the Hacktoberfest Open-Source AI Challenge with a .NET backend and vanilla web frontend. The application operates entirely offline to ensure user privacy and functionality in areas without cellular service. Developers overcame initial challenges by implementing strict prompt constraints to ensure the AI generates real-world tasks instead of fictional narratives.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

Guide automates real estate lead capture from web forms to CRM

A technical guide details automating lead capture for real estate websites. It describes using Contact Form 7 and a plugin to send inquiries directly into a CRM system. The process eliminates manual data entry, which is inefficient for the high-volume, time-sensitive nature of property inquiries. The setup involves configuring an API endpoint and mapping form fields to corresponding CRM lead fields. This automation ensures leads with structured data are created and can be immediately routed to the appropriate sales agent.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

New app uses local AI to suggest brief outdoor activities during work breaks.

A developer has created an app called Pocket Field Trip for the Hacktoberfest Open-Source AI Challenge. The app, launched on October 10, 2026, suggests short outdoor activities tailored to a user's available time, movement preference, and desired break type. It uses a local, open-source AI model to select an activity from a small set of options and then prompts the user to step away from the screen. The app operates entirely locally on a user's device and does not require a public server. Its developer notes the project is an initial version with limited activity coverage, and broader testing remains for future work.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

Benchmark Shows OCR Errors Often Stem from Reading Order, Not Image Quality

A developer ran a benchmark on September 30 to test 14 different image preprocessing methods for OCR on Chinese text. The test used the ImgIng service on different tiers and compared it to tesseract.js. Results showed that a high character error rate was often due to the text columns being read in the wrong sequence, not poor image quality. The issue was confirmed by a secondary metric that ignores character order, and rotating the text correctly resolved the errors.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

Home security system flagged own photo app as attack due to 404 burst

A homelab user discovered on September 21 that their CrowdSec security system had flagged a single local alert. The alert was triggered by their own phone's photo-backup app requesting thumbnails for deleted photos. A burst of 11 requests returning 404 errors within four seconds resembled probing behavior to the detector. The investigation revealed that high metrics counts originated from shared community blocklists, not actual attacks on the local network. This highlighted the importance of distinguishing between local alerts and pre-emptive blocklist data.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

Gartner Warns AI Risk Requires Double Investment, Survey Shows Gaps

Gartner advises companies to spend at least twice as much on derisking AI as on the AI tools themselves. A CEO survey found that while 75% are boosting tech spending, only 28% are increasing security budgets. Additionally, only 23% of leaders express high confidence in their generative AI outputs. The analysis suggests focusing on the software release process is a critical control point for managing AI risks related to cost, security, and trust.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

DevOps Engineer Demonstrates Kubernetes Persistent Storage and Azure IP Configuration

A DevOps engineer documented learning tasks involving Kubernetes and Azure. In one task, they created a PersistentVolume and a PersistentVolumeClaim in Kubernetes to provide storage for a web server pod that persists beyond the pod's lifecycle. In a separate Azure task, they attached an existing public IP to a virtual machine via its network interface configuration. The engineer successfully demonstrated that data written to the persistent storage remained accessible after deleting and recreating the pod. The tasks were part of a learning platform and highlighted the abstraction layers between pods and storage volumes, and between VMs and public IP addresses.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

Tesla Custom Wraps Require Specific Formatting to Avoid Common Errors

Tesla's Paint Shop feature allows owners to apply custom wrap designs to their vehicles using uploaded files. These wrap files must follow strict specifications, including specific image dimensions, file size limits, and correct formatting for the vehicle's exact model. Common issues include designs appearing misaligned on the car's 3D model or files not being recognized due to format violations. A developer has created a Python script to help owners validate their files against Tesla's published rules before attempting installation.

0 comments Read more at DEV Community
← NewerPage 542 of 3138Older →