Generative Matchmaking

Overview

Blog Post

https://daniel-deychakiwsky.github.io/generative-matchmaking

Running

Setup

This project uses poetry for python dependency management. Follow the instructions here for installation. Once, you've done that, clone this repository and run the following command to install the project's dependencies within the poetry-created virtual environment.

poetry install

Once you've done that, activate the project's python virtual environment.

poetry shell

Command Line Interface (CLI)

CLI commands assume repo root as current working directory.

The following command will list available CLI commands.

 python3 -m src.generative_matchmaking.cli

Generate User Profiles

This step can be skipped since these are already generated.

If you'd like to generate more on your own, run the following commands to set your OpenAI API key environment variable and generate synthetic dating user profiles.

export OPENAI_API_KEY="{{YOUR OPEN AI KEY}}"

python3 -m src.generative_matchmaking.cli \
generate-user-profiles \
--n-profiles 5 \
--model "gpt-4-0613" \
--max-tokens 5000 \
--temperature 1.05

Vectorizing User Profiles

This command will loop over all synthetically generated users and load them into a simple open source locally persistent vector database. See Chroma.

python3 -m src.generative_matchmaking.cli \
load-user-profile-collection \
--distance cosine

This command will query the vector database.

python3 -m src.generative_matchmaking.cli \
query-user-profile-collection \
--n-results 5 \
--verbose True \
--query-text \
"The user is seeking a man between the ages of 30 and 40,
who is at least 5'8\" and no taller than 6'3\". He should
have no children but wish to have them in the future. His
sexual orientation should be straight and he should be
someone who drinks sometimes, but does not smoke, use
marijuana, or other drugs. He should live an active
lifestyle. His ethnicity could be White Caucasian, East
Asian, or Hispanic/Latino and he should have liberal or
moderate political beliefs. Ideal occupations would be
within the Technology, Science, or Education industries.
He must speak English and have similar values, specifically
Honesty, Family, Responsibility, Respect, and Fairness.
His interests could include Technology, Outdoors, Cinema,
Reading, and Sports. The prospective partner should have
at least an undergraduate level education, with intentions
for long-term dating and a preference for monogamy."

This command deletes the loaded collection.

python3 -m src.generative_matchmaking.cli \
delete-user-profile-collection

Matchmaking

These are already generated but if you'd like to regenerate or generate more you can run these commands.

This command will run the matchmaking algorithm for a given user id.

python3 -m src.generative_matchmaking.cli \
find-matches \
--user-id "f0e35556-8760-41ae-b0f9-4c777c48b170" \
--n-matches 5

This command will do the same but for all user ids sequentially.

python3 -m src.generative_matchmaking.cli \
find-matches-for-all \
--n-matches 25

User Interface

To visualize the matchmaking we hacked together a user interface that will show a random selection of 10 user profiles and their matches. To run the user interface run the following commands.

This command will install node package manager (npm).

brew install npm

This command will change the current working directory to the ui directory.

cd ui

This command will install dependencies and run the local dev server which will open your default browser with the user interface. Everything this command is run a set of 10 random users will be selected and loaded. Feel free to run multiple times it to see different users.

npm install && npm run dev

---

This project was generated using the wolt-python-package-cookiecutter template.