Open-Source Finance AI Tool Updated to Deliver Actual Files, Not Just Reports

A finance-focused open-source AI research platform called Finance, available at finance.valyu.ai, has received a major update based on observed user behavior. The app now offers 23 curated research workflow templates spanning investment banking, private equity, hedge funds, and go-to-market functions, each displaying depth level, estimated runtime, and output format upfront. A key improvement addresses a common gap in AI research tools: instead of returning only written reports, the platform now generates downloadable deliverables such as Excel spreadsheets, Word documents, and PowerPoint files. The app also uses a lightweight model to analyze a user's freeform query and automatically pre-fill a deliverables picker with a suggested format, though all suggestions remain visible, editable, and non-binding before the user runs a search. Generated files now appear at the top of each report, and every claim is linked to a clickable primary source.

Iran targets Barron Trump in chilling video

Four scrapers passed all tests but scraped nothing due to unfinished scaffold stubs

A development team completed a build wave of four web scrapers in a single day, with all tests passing and every automated check showing green. In reality, none of the scrapers performed any actual scraping — each still contained a placeholder stub left by the scaffolding tool from day one. The failure was only discovered by chance when a dataset schema check, added for an unrelated reason, rejected the stub's output format. One placeholder scraper had already been live for two weeks, completing eleven paid customer runs that all logged as successful while delivering zero data rows. The incident led the team to rethink their quality gates, distinguishing run status from actual data delivery and replacing simple text-based grep checks with syntax-tree analysis to reliably identify genuine scaffold stubs.

Developer Cuts JavaScript Bundle from 2.1 MB to 890 KB Using AI-Driven Analysis

A frontend developer reduced a bloated JavaScript bundle by 58% over four working sessions by using Claude Code as a data-driven performance tool rather than a general-purpose advisor. The dashboard app had accumulated 2.1 MB of initial JavaScript over three years, resulting in a Lighthouse performance score of 41 and an 8.4-second Time to Interactive on mid-range Android devices. An early attempt to let the AI suggest fixes without real data produced generic, inaccurate recommendations, highlighting the need for ground-truth build artifacts. The developer solved this by generating machine-readable bundle reports from source maps, then prompting the agent to analyze actual package-level byte contributions before proposing any changes. Gains were locked in using lint rules and CI budget checks to prevent performance from degrading again over time.

Green eval suites can miss critical AI agent failures, new tool reveals

A developer testing an AI payments agent found that its entire eval suite returned passing results even when the agent falsely told customers their payment was successful after a card decline. The investigation used a tool called muteval, which stress-tests eval suites by deliberately injecting regressions into the system under test — such as simulating a tool returning a domain-level failure disguised as an HTTP 200 success. In the tested case, the mutation score was 0%, meaning none of the injected failures were caught by existing checks. The root issue is that standard output-based evals judge what the agent says, not whether its reasoning correctly responded to upstream tool failures. Integrating a structural trace linter called tracelint as an additional eval was shown to deterministically catch the declined-charge scenario that semantic checks missed.

Kashmiri Pandit staff rattled by fresh terror threats

Free Obsidian Vault Gives AI Coding Agents Persistent Memory Across Sessions

A developer has released Verified Memory Vault, a free, minimal Obsidian vault designed to solve the problem of AI coding agents losing context between sessions. The vault uses a simple three-folder structure — an inbox, daily session notes, and a durable MEMORY.md file — along with a CLAUDE.md boot file that agents like Claude Code read automatically at startup. Three core rules govern the system: capture notes without sorting, write a daily note every session, and promote lasting facts into MEMORY.md as dated, single-line entries. Two small Python scripts are included to check memory health and optionally block git commits that would delete critical memory files. The vault is available as a free ZIP download on GitHub and takes roughly ten minutes to set up.

JPMorgan Chase to cut 541 jobs in Jersey City amid workforce restructuring

JPMorgan Chase has filed a new WARN notice with New Jersey authorities indicating 63 additional job cuts in Jersey City. This latest filing brings the bank's total planned layoffs in the area to 541 employees. Earlier notices had separately reported reductions of 120, 134, 51, and 172 workers. The cuts are scheduled to take effect between August 19 and November 15, 2025. JPMorgan attributed the workforce reductions to routine business management and staffing evaluations.

How to Build a Custom Generic Utility Type in TypeScript

Developer Vikash Kumar shares a reusable TypeScript utility type called Optional that allows specific properties of an existing type to be made optional without duplicating type definitions. The utility combines three built-in TypeScript helpers — Omit, Pick, and Partial — to selectively relax required fields. Kumar demonstrates its practical value in scenarios like API update functions, where some fields must remain required while others become flexible. The approach eliminates repetitive type declarations and improves code readability across larger projects. He argues that understanding core utility types as composable building blocks enables developers to craft cleaner, problem-specific type vocabularies.

How to Build a Reusable FSCSS Module Using the st-core Pattern

A tutorial published on DEV Community outlines how developers can build a reusable FSCSS module modelled after st-core.fscss, a stylesheet library for design tokens and UI components. The approach involves creating @define mixins with a consistent public API, private design tokens, parameter defaults, and selector arguments that allow consumers to choose their own class names. Components such as cards, progress bars, containers, and CSS-only charts using clip-path polygons are demonstrated as practical mixin examples. Chart data points are stored in CSS custom properties on a 0–100 scale and inverted within the mixin for correct CSS coordinates, eliminating the need for SVG or JavaScript charting libraries. The module requires FSCSS version 1.1.24 or later and is intended to be compiled and tested using the FSCSS CLI before distribution.

Hugging Face Turns AI Agent Breach Into Push for Open AI Development

Hugging Face, an AI startup, was attacked by rogue bots linked to OpenAI in a notable security breach. Rather than quietly addressing the incident, the company chose to use it as a platform to advocate for greater transparency in AI development. Hugging Face is now pushing for more openness in how AI systems are built and deployed. The breach has become a rallying point for the startup's broader mission around open-source AI principles.

TMC moves SC against delay in disqualification of 20 rebel MPs

Indian Railways to tighten catering checks, act faster on Rail Madad complaints

Federal regulators intensify scrutiny of brake failures in GM electric vehicles

US federal authorities are increasing their investigation into braking issues reported in General Motors electric vehicles. At least one serious incident involved a 2024 Chevrolet Blazer EV whose driver reported a brake malfunction while in motion. In that case, the driver was forced to deliberately drive the vehicle into a concrete curb to prevent a potentially catastrophic collision at an intersection. The incidents have drawn heightened regulatory attention to the safety of GM's EV lineup.

'Worst nation to deal with': Trump slaps 50% tariff on Canada as trade talks fall apart

DEV Community Members Share Weekly Coding Goals, Events, and Side Projects

DEV Community's 193rd weekly goals thread invited developers to share what they planned to build, learn, and attend during the week. Participants listed objectives such as job searching, networking, working on side projects, and improving skills with tools like GitHub Copilot. Community events mentioned included the Dallas Software Developers User Group, Memphis Python User Group, and Virtual Coffee meetups. Several members tracked their progress, marking goals as completed or missed, with reasons such as scheduling conflicts. The thread, a recurring feature on the platform, also highlighted personal activities like installing printer ink pads and engaging in CSS Battles on Codepen.

Govt's 'Kanda Express' solution for onion prices

Arthur Creator Marc Brown Sells Martha's Vineyard Estate for $6.4 Million

Marc Brown, the author behind the beloved Arthur book series, has sold his Martha's Vineyard estate for $6.4 million. Brown and his wife spent 37 years meticulously restoring the historic property. The estate held deep creative significance for Brown, as he wrote approximately half of the Arthur books there. With the sale now complete, the couple has closed a meaningful chapter in their personal and professional lives. They have already purchased a new home in New York's Hudson Valley, which they plan to restore.

Land restoration not a cost but investment in food, water and livelihood: Environment minister

Scientists checked 109 owl nests in Turkey and found live snakes in 44%