AI / LLM application
Document Explainer
Explains confusing official US letters in plain language — and in your language.
- Python
- Flask
- Anthropic Claude (claude-sonnet-5)
- pytest
The problem
Official letters — from the DMV, IRS, a bank, a landlord, or an insurer — are full of jargon and deadlines. People who are new to the US often can't tell an urgent notice from junk mail, and end up relying on second-hand interpretation.
Who it's for
People new to the US and the systems around them — and honestly, anyone who wants a dense official letter turned into something they can act on in 30 seconds.
What it does
A web app where you paste the text — or upload a photo or PDF — of a letter and get four clear things back: what it is, what you need to do (with any deadlines), what happens if you ignore it, and the whole summary translated into your language.
Key decisions
- Forced the AI output into four fixed sections, so the answer is always scannable and actionable — never a wall of text.
- Added a clearly-marked demo mode that runs with no API key, so the interface can be tested instantly and safely.
- Built in translation and a multilingual UI, so the tool meets people in their own language.
- Kept a firm boundary in the copy: it explains what a document says — it is not legal, medical, or financial advice.
What I learned
How to turn an LLM into a dependable product feature instead of a demo — shaping the prompt for consistent structure, handling the no-key case gracefully, and writing tests around AI output.
Where it goes next
Save explanation history, support more document types, and add a mobile-friendly camera-capture flow.