Regex couldn't sort my sales emails. A yes/no judgment model could

Описание и применение

Регулярные правила не справились с сортировкой писем, поэтому автор использовал модель Jev для принятия решений о каждом письме, что сократило количество сообщений в почтовом ящике с 92 до 2.

Задача

Gmail filters and regex failed to separate sales pitches, auto‑replies, newsletters, and bounces from genuine replies, leaving the inbox cluttered.

Результат

The automation reduced inbox clutter from 92 to 2 messages, correctly filtering out pitches, auto‑replies, and bounces while preserving necessary replies.

Рабочий процесс

• 1. Identify email types that can be handled by code (bounces, invoices, payments, contracts).
• 2. For each remaining email, send a single request to TypeSafe AI’s Jev system with five noul questions: human, needReply, counterSales, meetingRequest, stopRequest.
• 3. Run each request twice and only act when both runs agree on the same verdict; otherwise flag for manual review.
• 4. Apply threshold lines per question (e.g., human yes 0.5, no 0.3) to decide actions.
• 5. Cache judged message IDs to avoid double‑charging.
• 6. Schedule the process to run every five minutes.

Доказательства

On October 2, 2026, it took the inbox from 92 messages to 2.

Источник

I used Jev from TypeSafe AI. It's what TypeSafe calls a System One model...

Источник

On October 2, 2026: about 70 pitches and invites moved to "no reply needed"

Источник