SShortSingh.
Back to feed

Developer Builds AI Finance App Where Model Can Propose But Never Touch Your Data

0
·2 views

A developer has built Sable, a React Native personal finance app designed so that all financial data is stored locally on-device in SQLite, with no cloud backend. The app uses OpenAI function calling, but every AI-suggested action — such as logging a payment — is rendered as a 'Review & Confirm' card that only executes when the user manually approves it. This architecture ensures an AI hallucination can produce, at worst, a dismissible card rather than an erroneous ledger entry. Additional features include serialized database writes, an offline-first design, and a daily on-device RAG-powered Morning Briefing that summarizes spending without sending any financial data off the device. The developer argues this 'propose/confirm' pattern is broadly applicable to enterprise domains like healthcare and legal, where AI assistance is valuable but full AI authority over data is unacceptable.

Read the full story at DEV Community

This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)

Log in to join the discussion and vote.

Log in

Related stories

0
ProgrammingDEV Community ·

Developer Builds Lightweight Wiki and Database for Steal an Egg Game

A developer shared on DEV Community how they built a fast, minimalist wiki and database for the Steal an Egg gaming community. The project was motivated by frustration with existing resources cluttered with heavy ads, slow load times, and SEO spam. The site was built using a modern framework such as Next.js or Vue, styled with Tailwind CSS, and uses MDX for structured game content including egg stats and drop rates. Data is organized through JSON or database layers and optimized for Core Web Vitals and search engine performance, with fast client-side filtering for in-game lookups. The developer plans to add community contributions and automated data syncing, and has invited feedback and pull requests from the community.

0
ProgrammingDEV Community ·

Free App YoFi Unifies Income, Expenses, and Net Worth for Indie Devs

A developer has launched YoFi, a free financial dashboard designed for indie hackers, freelancers, and digital creators whose income is spread across platforms like Stripe, RevenueCat, and Gumroad. The app consolidates online revenue alongside personal finances, including expenses, savings, stocks, and crypto, to give users a single view of their net worth. YoFi also supports budgeting, subscription tracking, OCR-based receipt entry, and over 160 fiat currencies, catering to remote workers and digital nomads with multi-currency needs. The tool is currently available on Android and accessible via its web link, yofiapp.vercel.app. The creator built YoFi to address the common frustration of toggling between multiple dashboards just to answer basic questions about one's overall financial position.

0
ProgrammingDEV Community ·

Wrong SHA-1 Fingerprint in Play Console Silently Breaks Google Sign-In on Android

Developers using Google Sign-In on Android may find the feature works in debug builds but silently fails in production apps distributed via Google Play. The root cause is a fingerprint mismatch: Play App Signing replaces the developer's upload key with its own certificate before delivering the app to users, meaning the SHA-1 registered in Google Cloud Console must match Play's signing certificate, not the upload key. When the fingerprints don't match, Android's Credential Manager returns a generic 'cancelled' result with no error message, making the bug nearly impossible to diagnose. Developers who have opted into Google's Quantum-ready app signing beta face an added complication, as the SHA-1 buttons shown on Play Console's App Signing page no longer reflect the actual certificate signing the app. The fix requires downloading the deployment_cert.der file from Play Console, extracting its SHA-1 fingerprint, and registering it as an additional OAuth client in Google Cloud Console.

0
ProgrammingDEV Community ·

How Anthropic's Internal Fix Became the AI Industry's Universal Standard

The Model Context Protocol (MCP) began in mid-2024 when Anthropic engineers David Soria Parra, Justin Spahr-Summers, and Mahesh Murag built an open standard to let AI assistants connect seamlessly with external tools, data sources, and development environments. The protocol addressed the so-called N×M problem, which required a unique custom connector for every possible AI-tool combination. Major spec updates in March and November 2025 introduced OAuth 2.1 authentication, real-time data transport, asynchronous operations, and a community-driven server registry. Adoption expanded rapidly across the industry, with OpenAI, Google DeepMind, and Microsoft all embracing MCP, and Anthropic eventually donating the project to the Linux Foundation's newly formed Agentic AI Foundation to make it vendor-neutral. By mid-2026, MCP had become ubiquitous across leading AI platforms, with its two largest SDKs recording over 470 million downloads in a single 30-day period.