Agents for Humans: an agent that refuses to guess what your letter says

Description and use

Plainletter is an AI agent that automatically extracts, verifies, translates, and drafts responses for foreign-language letters, reducing errors and speeding up help desk operations.

Problem

People receive foreign-language letters they cannot read and rely on free help desks to interpret them, but volunteers may miss details or misinterpret information, leading to errors in handling deadlines, amounts, and references.

Outcome

Plainletter accurately extracts and verifies information from foreign-language letters, reducing errors such as misread amounts or dates. It supports multiple languages, enforces strict grounding of facts, and integrates cost control by limiting Bedrock calls.

Workflow

• 1. Volunteer photographs or uploads the letter.
• 2. Plainletter extracts sender, letter type, amounts, reference numbers, and deadlines, marking each on the letter.
• 3. It verifies each extracted fact against the letter text, refusing any ungrounded data.
• 4. It translates the letter side-by-side into the visitor’s language and counts days left.
• 5. It retrieves official routes from a knowledge base via a tool call and drafts a reply if appropriate.
• 6. It prints a desk card with a calendar reminder.
• 7. The volunteer reviews the output and handles any unresolved items.

Evidence

Plainletter names the sender and the kind of letter, pulls out every amount, reference number and deadline, and marks each one on the letter with a numeral in the margin.

Source

For every date, amount and reference, it checks that the cited passage really stands in the letter and that the value follows from it.

Source

One reading is six Bedrock calls and costs about USD 0.107.

Source