In developmentTax services business
Dari & Pashto AI Tax-Intake Assistant
A tax business serves Dari- and Pashto-speaking clients who struggle with English intake forms — and no off-the-shelf tool speaks their languages.
roleFounder-engineer: AI workflow design, APIs, and integration
timeline2026 — present
statusin development
- Google Gemini
- WhatsApp channel
- TypeScript
- Structured outputs
- MCP server
- Reusable APIs
// problem
Tax intake requires collecting precise personal and financial information from clients who communicate most comfortably in Dari or Pashto. Manual translation makes every intake slow and error-prone.
// users
Dari- and Pashto-speaking tax clients on WhatsApp; the tax business receiving structured intake data.
// constraints
Intake conversations must produce structured, verifiable data — not chat transcripts. The languages are low-resource, so prompt and workflow design carry the quality burden.
// solution
A Gemini-powered conversational intake assistant designed for a WhatsApp communication channel. It conducts the intake in Dari or Pashto, structures the collected information, and transfers it to the tax business.
// architecture
Gemini language models behind reusable APIs, with an MCP server exposing the multilingual capability so other applications and businesses can consume it as a service.
// my role
- Multilingual conversation and prompt design for Dari and Pashto
- Structured-output pipeline from conversation to intake record
- Reusable API surface and MCP server design
- Customer discovery with the tax business
// status & outcomes
In active development. Intended to become a reusable service for organizations that need Dari and Pashto communication.
// lessons
For low-resource languages, evaluation with native speakers beats any benchmark — I speak both target languages, which closes the loop most teams can't.