galeone / fitsleepinsights

Custom dashboard for Fitbit users. RAG support (gemini-based) for chatting with your data and getting insights about your fitness activities and sleep data

Fit Sleep Insights

Web application that connects with your Fitbit account and tries to do explainable predictions of your sleep quality - using all the available data.

• What makes you sleep well?
• What are the activities you should perform and when?
• What you should avoid?
• What increases your deep sleep?

These are some of the question this project tries to answer.

The goal is to understand what behaviors/activities are good/bad for your sleep. People are different, and what makes a person sleep well could make some one else sleep a nightmare. For this reason, the prediction are made per user - no data is aggregated.

Technicalities

The application is made of different parts

• A Fitbit API client. Folder fitbit
• A storage. The storage requires a dedicated setup (we use TimescaleDB/PostgreSQL). Folder storage.
• The predictive models
• The web application

The web application, together with the client and the database asks the user permission on the Fitbit account.

Docker-Compose

You can chose to configure the application via the docker-compose.yaml file, and executed it trough docker compose - or you can configure (without depending on Docker) the various components of the application locally.

Below you can find the steps to setup the application locally.

Database setup

1. Install PostgreSQL

2. Create the user and the database. Note: on MacoOS skip the sudo -i -u postgres part of the commands.

sudo -i -u postgres createuser -U postgres fitsleepinsights
sudo -i -u postgres createdb -U postgres fitsleepinsights fitsleepinsights

3. Grant the create privilege (required from PostgreSQL 15+), and make it superuser (required for the vector extension):

psql -U postgres -d fitsleepinsights -c "GRANT USAGE, CREATE ON SCHEMA public TO fitsleepinsights;"
psql -U postgres -d fitsleepinsights -c "ALTER USER fitsleepinsights WITH SUPERUSER;"

4. Install pgvector: https://github.com/pgvector/pgvector

Done. On startup the application creates the schema if not present.

Configuration

Place a .env file in the runtime path. This is the content:

# Register the application
# https://dev.fitbit.com/build/reference/web-api/developer-guide/getting-started/
# for obtainign these values.
FITBIT_CLIENT_ID=""
FITBIT_CLIENT_SECRET=""
FITBIT_REDIRECT_URL=""

# Use the same values used in the Database setup section
DB_HOST=localhost
DB_PORT=5432
DB_USER=fitsleepinsights
DB_PASS=fitsleepinsights
DB_NAME=fitsleepinsights


# Web application settings
DOMAIN=fitsleepinsights.app
PORT=8989

# Vertex AI
VAI_LOCATION="europe-west6"
VAI_PROJECT_ID="project id"
VAI_SERVICE_ACCOUNT_KEY="full path"

Running

Yup, running is important. There are 2 ways of running the application, depending on how you configured it.

If you installed the database and configured the .env file you need simply to run it.

go run main.go

If you configured the docker-compose.yaml file, than you can:

./run.sh up
# and to stop it
./run.sh down

Or you can install it with go install and execute fitsleepinsights.