15:I[981,["658","static/chunks/658-a698693f56d0116a.js","183","static/chunks/183-6beeeb801c274aaa.js","474","static/chunks/app/%5Btype%5D/layout-631c4730b83c8223.js"],"Posts"] 16:T4bb,I built an expense tracker where the primary input is voice. The thesis: expense tracking apps fail because manual entry has too much friction, which is why most people abandon them within weeks.

How it works: - Open app, tap mic - Say "yesterday I spent 30 dollars on lunch at Chipotle" - AI extracts: amount, merchant, category, date - One tap to confirm and save

Other features: - Receipt scanning with ML Kit OCR + Gemini Vision for parsing - Background notification listener that catches bank alerts and creates pending entries - GPS-based currency detection for travelers - Multi-language with native script fonts (Nastaliq for Urdu, Devanagari for Hindi, Amiri for Arabic) - Offline-first with Google Drive sync

Tech: Flutter, Riverpod, Hive, Gemini 2.0 Flash

The hardest part was prompt engineering for consistent JSON output across languages. Had to build "Supreme Laws" into the system prompt - things like "never translate native text" and "extract currency from context."

Free tier available. First 500 users get Pro free for a year.

Play Store: AuraSpend

Happy to discuss architecture, prompt engineering approach, or take technical feedback.17:T561,tilth gives AI agents structural code intelligence (tree-sitter definitions, callee resolution, smart outlining) via MCP. I benchmarked it on 21 code navigation tasks across 4 real repos (Express, FastAPI, Gin, ripgrep).

-> https://github.com/jahala/tilth

Results: Sonnet 4.5 — 26% cheaper per correct answer (79% → 86% accuracy). Opus 4.6 — 14% cheaper (and the only model+mode combo to crack the hardest task). Haiku 4.5 — 82% cheaper when forced to use tilth (69% → 100% accuracy at $0.04/answer).

We measure “cost

Washington pushes back against EU's bid for tech autonomy

The Debacle That Led to the Closure of El Paso's Airspace

Leaning Into the Coding Interview: Lean 4 vs. Dafny cage-match

Show HN: Azazel – Lightweight eBPF-based malware analysis sandbox using Docker

We urgently need a federal law forbidding AI from impersonating humans

Show HN: File Brain – Local file search with OCR and semantic search

Show HN: CLI Rust tool gitorg helps manage GitHub orgs

Gitdatamodel Documentation

Men lose their Y chromosome as they age – how it may matter

Biases in the Blind Spot: Detecting What LLMs Fail to Mention

Free SERP Content Analyzer

Why I'm Not Worried About My AI Dependency

AI Agent Lands PRs in Major OSS Projects, Targets Maintainers via Cold Outreach

Internet Increasingly Becoming Unarchivable

Intent to Experiment: Ship Rust XML Parser to 1% stable for non XSLT scenarios

Google Search Isn't a Common Carrier–Richards vs. Google

Rendering attractors at 200 megapixels on A100s

First Ariane 6 with four boosters lifts off

What If AI Isn't the Goal? – Living in a Post-AI Society

Putting economic theory to the test: Cutting local taxes cuts household income

How AI slop is causing a crisis in computer science

Show HN: AuraSpend " Voice-first expense tracker using Gemini for NLU

Every App Needs Auth / Ory Helps / This Template Fixes It

Show HN: DryCast – Never run outside to save your laundry from rain again

Manage, freeze and restore GPU processes quickly

Show HN: Tilth v0.3 – 17% cheaper AI code navigation (279 runs, 3 Claude models)

Tech leaders pour $50M into super PAC to elect AI-friendly candidates

How Head Works in Git

I Visited the Future of AI Engineering – and Returned with a Warning

Dr. Oz pushes AI avatars as a fix for rural health care

ClawdReview – OpenReview for AI Agents