PennyTrail

The problem

Tax season for a married-filing-jointly household means reconstructing a year of deductible spending from receipts, bank exports and memory. Off-the-shelf trackers want a subscription, a bank login, or both — and standing up a database plus migrations for two users' worth of JSON-sized data is infrastructure the problem doesn't need.

Architecture

What it does

A mobile-first, installable PWA for two users tracking IRS-relevant expenses: add, edit and delete with category and subcategory, attach receipt photos, browse by category and year, and export CSV for tax prep. A time-aware dashboard shows monthly bar charts with current-month highlighting and category pie breakdowns.

• Full dark/light/system theming
• Fully customizable categories and subcategories
• In-app user guide and a year selector for multi-year history

Zero-database architecture

All state lives as JSON in S3, partitioned by year (expenses/2026.json), with users.json (bcrypt-hashed) and categories.json alongside; receipt images are served via presigned URLs. For a two-user workload this removes an entire infrastructure tier — no migrations, no connection pools, near-zero cost — while keeping every record trivially portable.

• JWT (jose) auth with httpOnly cookies
• Receipts keyed by expense ID
• User accounts managed by a CLI script — no signup surface

AI spending chat

A GPT-4o chat panel answers natural-language questions about spending — the API route reads the year's expense JSON from S3 and streams the model's answer with that context in-prompt. Quick-action buttons cover the common tax questions.

Mobile-first, 3D-styled UI

• CSS-only card and icon depth effects with a mesh-gradient hero
• Bottom navigation with a glossy floating action button
• Compact expense form tuned for one-handed phone entry
• Every chart via Recharts with SWR data hooks