A grounded AI assistant sitting on top of a purpose-built cross-border tax knowledge base covering 147 countries. It answers withholding-tax and treaty questions by retrieving from verified data rather than generating from memory, and shows the corridor data behind every answer.
The problem
Cross-border withholding tax is exactly the kind of question a general chatbot answers confidently and wrongly. The real rates live in treaty tables that differ per country pair, per income type, and change over time — so an answer without a verifiable source behind it is worse than no answer.
What I built
Live and publicly reachable, running as a containerised app on a free hosting tier with a self-healing database bootstrap that re-seeds itself on integrity failure or version change.
How it works
Rather than trusting a model's recall, the system ingests official treaty withholding data into a local knowledge base, marks each country corridor with how well-sourced it is, and only lets the assistant answer from what it can retrieve. Ingesting the OECD treaty withholding dataset added tens of thousands of verified rows and promoted roughly six hundred non-US corridors from single-source to verified. A global explorer lets you browse all 147 countries directly, and answers export to a PDF memo.
Highlights
- 147-country sortable tax explorer built from a real territories dataset
- 34,819 OECD treaty withholding rows ingested and reconciled
- ~600 country corridors promoted from single-source to verified
- Answers are grounded in retrieved data and exportable as a PDF memo
- Holding-structure optimiser and follow-and-alert on corridors you care about
- Self-healing container bootstrap re-seeds the database on integrity failure
Let's build yours.
Tell me what you're trying to ship — you'll get a scoped plan and a straight answer.