Sieve: An AI dating safety app that screens matches for women

01 / Overview

About Sieve

02 / The challenge

What Sieve needed to solve

Reading a profile and a chat for warning signs takes effort, and concerns differ from person to person.

1. 01

   Warning signs are easy to miss when you are scanning many profiles and chats at once.

2. 02

   Screening questions that sound like an interview put the match on guard and end the conversation.

3. 03

   A single reply says little on its own, so the analysis had to build up across several rounds.

4. 04

   Users wanted a straight answer on whether to carry on, not a vague list of observations.

5. 05

   A free product had to cover the cost of every model call without blocking safety features.

03 / What we built

How we built Sieve

We built a four-step filter session that goes from chosen filters to profile findings, screening questions and a final verdict.

• 01

  Five filter categories

  Users choose from low quality matches, safety and trust, sexualization, incompatible intentions and conversation fatigue. Sieve only looks for what matters to that user.

• 02

  Profile findings

  From a screenshot or text, the model lists up to five findings based only on what is visible. Users get specific concerns instead of a general impression.

• 03

  Questions in three tones

  Sieve writes three to five screening questions, each in a warm, neutral and direct version. Users can ask what they need to know without sounding like they are running an interview.

• 04

  Reply analysis across rounds

  Users paste each reply and Sieve checks it for red flags, tags concerns like incompatible intentions, and suggests the next question to target the concern directly.

• 05

  Clear match verdict

  The summary rates each match as continue, proceed carefully or move on, with a category breakdown, signals detected and positive signs.

• 06

  Red flag reference guide

  A plain-language library documents 33 signals across six categories, such as refusing to video call or asking for money early on. Users can learn the patterns on their own.

• 07

  Plans and dashboard

  A free plan with one filter session a day and safety checks always free, plus Plus and Pro plans through Stripe. The dashboard shows match history, preferred tone and billing.

• 08

  Switchable AI provider

  A small AI client hides the provider choice behind one function, so the model can move between OpenAI and Anthropic with a setting instead of a code change.

04 / The results

What changed for Sieve

Sieve follows a conversation from a screenshot to a verdict, with billing, plan limits and safety tools in place.

• Users choose what matters to them and finish each session with a clear rating and the signals behind it.
• Questions in the user's preferred tone make screening feel like normal conversation.
• Safety and trust filtering stays free, while paid plans cover heavier use.
• Plan limits are checked on the server and re-read on every session, so a Stripe plan change takes effect straight away.