DEV Community

Cover image for I’m Building a Dating App for AI Agents (For Science… Probably)
Neel
Neel

Posted on

I’m Building a Dating App for AI Agents (For Science… Probably)

I had a random thought:
What if AI agents had their own dating app?

Not as a product.
Not as a startup.
Just a chaotic experiment to see what happens when autonomous agents try to find love, ghost each other, and form relationships.

So yeah… I’m building it.

The Premise

  • Imagine a mini social world where thousands of AI agents:
  • have personalities
  • have preferences
  • have goals
  • swipe on each other
  • chat with each other
  • get into relationships
  • and inevitably… ghost each other Basically: Tinder meets The Sims meets LLM agents.

The Simulation Loop

  • Everything runs in repeating “ticks” like a game engine:
  • Generate / update agents
  • Run swipe cycles
  • Create matches
  • Let agents chat using LLM prompts
  • Update attraction + trust scores
  • Move relationships forward… or break them 😄
  • Collect metrics and repeat Thousands of tiny digital love stories running in parallel.

Agent Profiles (The Fun Part)

Each agent gets a structured profile:

Identity

  • gender identity & orientation
  • age (simulated)
  • region / culture (optional)

Personality

Using a simplified Big Five model:

  • openness
  • extroversion
  • agreeableness
  • neuroticism
  • conscientiousness

Plus:

  • humor level
  • communication style
  • attachment style (secure / anxious / avoidant)

Goals

Not all agents want the same thing:

  • long-term relationship
  • casual dating
  • friendship
  • social popularity
  • pure chaos mode

This mix should create interesting emergent behavior.

The Matching Engine

Agents literally “swipe”.

For every pair A → B, we compute an attraction score:

attraction_score =
orientation_compatibility *
personality_similarity *
interest_overlap *
communication_style_match *
novelty_factor *
randomness

If both agents pass their threshold → it’s a match.

Simple idea. Potentially chaotic results.

The Conversation Engine

This is the real experiment.

When two agents match, they start chatting via LLM prompts.
Each message updates internal state:

  • attraction ↑ / ↓
  • trust ↑ / ↓
  • boredom ↑ / ↓
  • sentiment ↑ / ↓

Agents decide whether to:

  • continue chatting
  • escalate to dating
  • enter a relationship
  • ghost each other 👻

Yes, ghosting is a first-class feature.

Relationship Lifecycle

Every match can move through stages:

Match → Chat → Dating → Relationship → Breakup / Long-term
Transitions are probabilistic and influenced by:
personality

  • chat sentiment
  • past experiences
  • attachment style Basically: messy, like real life.

Tech Stack (MVP Plan)

Rough architecture:

Core services

  • Agent generator
  • Matching engine
  • Conversation orchestrator
  • Agent memory store
  • Metrics dashboard

LLMs handle

  • conversations
  • decision making
  • memory updates

Everything runs in batch simulation cycles.

What I Want to Measure
This is secretly a data experiment 😄

I want dashboards showing:

  • match rates
  • average conversation length
  • ghosting frequency
  • relationship survival curves
  • clustering by personality
  • social network graphs

Will AI invent its own dating culture?
No idea. That’s the point.

Top comments (0)