7,044 died as influenza swept Delhi a century ago

Lenders forming 'union' against Chandra's haircut

Developer shares 10 critical checks before trusting AI-generated code

A software developer writing for DEV Community has outlined ten practices they follow to responsibly use AI coding tools without blindly relying on them. Key concerns include verifying unfamiliar terminal commands, auditing third-party packages for legitimacy, and never exposing secret credentials through AI prompts. The developer also warns that AI-generated authentication logic and database migration scripts may appear functional but carry serious security or data-loss risks in production environments. They emphasize that developers must fully understand any AI-written code before merging it, since untested or opaque logic can introduce hard-to-trace bugs later.

Bombay HC closes plea after Maharashtra extends Marathi learning deadline by one year

The Bombay High Court has disposed of a petition that challenged Maharashtra's rule requiring cab and auto-rickshaw drivers to learn Marathi. The petition had claimed the language requirement violated the fundamental rights of drivers. The court's decision to close the case came after the state government announced a one-year extension to the compliance deadline. The extension gives drivers additional time to acquire a working knowledge of Marathi before the rule takes effect.

India must cut reliance on foreign tech: M2P Fintech

How Businesses Can Cut Proxy Costs by Up to 60% Without Losing Performance

Proxy services are a major operational expense for businesses, with typical monthly spending on residential proxies ranging from $500 to $5,000 depending on bandwidth usage and provider. The primary cost driver is bandwidth consumption, as most providers charge per gigabyte, while hidden costs also arise from failed requests, geographic diversity, and concurrent connection needs. Experts suggest that the gap between an unoptimized and a well-optimized proxy strategy can translate to 40–60% in potential savings. Key optimization tactics include enabling gzip compression to cut payload sizes by up to 80%, fetching lightweight API endpoints instead of full web pages, and caching repeated requests to eliminate redundant traffic. One real-world example showed a SaaS company reducing monthly bandwidth from 150GB to 35GB — cutting costs from $450 to $105 — purely through request optimization, without switching providers.

How to Run a Local AI Chatbot on Your PC Without Risking Your Privacy

Large language models (LLMs) can now be installed and run directly on personal computers, offering users a capable AI assistant without relying on cloud services. Running a chatbot locally means your data stays on your own device, eliminating privacy concerns tied to third-party servers. This approach gives individuals greater control over their AI tools compared to using web-based platforms. The setup process has become increasingly accessible, making local AI assistants a practical option for everyday users.

‘Maybe he was calling to tell us about the flood’: A family waits for news from Nepal

Developer builds 9 browser-based puzzle solvers using classic algorithms, no servers needed

A developer has built a suite of nine puzzle and game solvers — including Sudoku, Minesweeper, and Tic-Tac-Toe — that run entirely in the browser without any backend, API calls, or machine-learning models. Each puzzle solver was designed to use a different classic computer science algorithm, such as constraint propagation, minimax search, and alpha-beta pruning. The project runs on vanilla JavaScript inside a Web Worker, completing computations within a few hundred milliseconds on the user's own device. A key design principle was keeping everything client-side, both to protect user privacy and to eliminate server costs. The developer notes that these constraints made classic algorithms the ideal choice, as they are fast, deterministic, and lightweight enough to ship as a browser script.

Last call from Nepal, then silence for Thane family

Free AI Quotas Come With Hidden Risks Developers Often Overlook

Developers using free AI model quotas or servers often test the model itself but neglect to evaluate the underlying hosting infrastructure, according to a practical guide published on DEV Community. Free tiers typically carry no SLA, no burst guarantees, and no data-boundary promises, making them unsuitable for certain production workloads. Key risk factors include hard latency requirements, bursty demand patterns, stateful workflows, restricted data policies, and limited observability when failures occur. The guide recommends scoring six infrastructure red flags and running a short automated probe script to measure availability, latency, and error rates before connecting a free endpoint to any agent. Developers are advised to assess the host and the model independently, since a high-quality model can still fail users when the hosting layer is unreliable.

FIR filed against Mamata over 'violation' of HC order

Developer Builds Parametric OpenSCAD Bayonet Hose Connector for Mobile AC Units

A developer has created a fully parametric bayonet connector in OpenSCAD designed to attach hoses to mobile air conditioning units and window exhaust kits. The design consists of three 3D-printable parts — a female connector, a male connector, and a female adapter — that lock together with a simple push-and-twist action, eliminating the need for threaded joints or tools. The model is published as part of an Air Conditioning Collection on MakerWorld, where users can customize key parameters such as hose diameter and fitting height through the platform's Parametric Model Maker. To aid understanding, the OpenSCAD file includes a 10-scene animation that illustrates how the bayonet mechanism works and demonstrates the model's parametric flexibility. The project also supports MakerWorld's multi-plate export feature and includes subtitle files, making it a comprehensive open-source hardware resource.

Key Lessons From Studying EKS Cluster Upgrades the Right Way

An SRE newcomer shared practical insights after deeply researching Amazon EKS cluster upgrades, moving beyond the common assumption that a version bump is a simple console click. A critical finding is that EKS control plane upgrades are irreversible, meaning thorough pre-upgrade testing is mandatory since rollback is not an option. Deprecated Kubernetes API versions pose a hidden risk, particularly through forgotten Helm charts and third-party tools, which can be caught using utilities like Pluto and kube-no-trouble before an upgrade proceeds. Add-ons such as CoreDNS, kube-proxy, and the VPC CNI plugin must be manually verified and upgraded alongside the control plane, as compatibility is not guaranteed automatically. Admission webhooks from tools like cert-manager or service meshes can also silently break cluster-wide API requests if not confirmed compatible with the new Kubernetes version ahead of time.

Viking textile fragments preserved by jewelry corrosion found in Denmark

Danish archaeologists have unearthed rare Viking-era textile remains near the town of Varde in Denmark. The wool and linen fabric fragments, approximately 1,300 years old, survived because corrosion from nearby jewelry helped preserve them. The discovery provides an uncommon window into how Viking women dressed during that period. Researchers plan to analyze the textile fibers and skeletal teeth to extract further historical details. Dating the textiles may also reveal whether the jewelry buried alongside them was newly acquired or passed down as an heirloom.

Russian strikes near Kyiv kill at least 27, including a teenager

At least 27 people have been killed in Russian strikes carried out near the Ukrainian capital Kyiv. Among the victims is a 14-year-old girl, according to the regional governor. The attacks prompted the evacuation of hundreds of residents from affected areas. The strikes targeted locations in the Kyiv region overnight, adding to the ongoing toll of the conflict in Ukraine.

Russian Strikes on Kyiv Kill 27, Damage 50 Residential Buildings

Russian strikes on Ukraine's capital Kyiv have killed at least 27 people and damaged around 50 residential buildings. The attack is part of an escalating pattern of long-range strikes exchanged between Russia and Ukraine in recent months. Both sides have intensified such strikes, resulting in widespread fires and rising civilian casualties. The frequency and scale of these attacks have pushed civilian death tolls to record levels.

Key Considerations for Managing Cisco ACE Load Balancers in Production

The Cisco ACE load balancer, available as the ACE 4710 appliance or as a module inside network hardware, sits between clients and backend servers acting as a reverse proxy or Layer 4 load balancer. Administrators must understand how virtual IPs, class maps, policy maps, server farms, and health probes interact as a connected chain rather than isolated settings. Traffic distribution methods such as weighted round-robin and least-connections serve different purposes, and choosing between them should be based on actual application traffic patterns rather than algorithm names alone. Health probes require careful selection, since a TCP probe only confirms a port is open, while an HTTP probe against a dedicated endpoint provides a more accurate signal about whether the application is truly ready to serve traffic. Routing design, slow-start for new servers, and connection state management are also highlighted as factors that directly affect load balancer reliability and must be considered alongside the core configuration.

Developer Builds Hybrid RAG System Combining FAISS Vector Search and BM25 Retrieval

An AI engineering practitioner has documented the development of a Hybrid Retrieval-Augmented Generation (RAG) system that merges FAISS semantic vector search with BM25 keyword search to improve document retrieval accuracy. The system splits source documents into chunks, converts them into vector embeddings for FAISS indexing, and simultaneously indexes them for BM25 term-based matching. Scores from both retrieval methods are normalized and combined through weighted ranking before the top results are passed as context to a large language model. The project also incorporates an agentic layer, exposing the retrieval pipeline as a callable tool so an AI agent can query the knowledge base before generating grounded responses. Qwen2.5-72B-Instruct was used as the underlying language model, with the system designed to flag queries when the knowledge base lacks sufficient information.

Glenn L. Martin National Wildlife Refuge Shields Migratory Birds Across 2,500 Acres

The Glenn L. Martin National Wildlife Refuge, established in 1954, was originally a private hunting preserve before being repurposed for wildlife conservation. Spanning over 2,500 acres, the refuge encompasses tidal marshes and wooded ridges that serve as critical habitat for migratory birds. The area provides seasonal shelter for waterfowl and nesting grounds for numerous bird species throughout the year. Beyond birds, the refuge also supports mammals, reptiles, and marine life, helping preserve the broader bay ecosystem. The site has grown considerably in size since its founding, strengthening its role as a key sanctuary for regional biodiversity.