A service to send me notifications about what I did on this day in years past.
For several years I have been writing handwritten descriptions of what I get up to each day. The idea is to keep track of my day to day activities, not just bigger events in my life. On any day, I can look back and see what I did on that date in previous years.
I need to make digital copies of the diary entries. Typing up years of handwritten entries would be too tedious and time consuming. So instead, I take scans of each diary entry. An example (blurred to hide details) is shown below.
Next I crop and label each image using the date the entry corresponds to. For example 20210624.jpg. If the handwritten entry spans multiple pages, append a letter to each (20210624A.jpg, 20210624B.jpg, etc.).
This solution is deployed using Google Cloud Platform, orchestrated using Terraform.
- Create a bucket in GCS.
- Manually upload the diary entries to the bucket.
- Deploy a Google Cloud Function that finds the diary entry in the bucket that matches today's date.
- Create a Cloud Scheduler to trigger a Pub/Sub message once a day. The Pub/Sub then triggers the GCF.
I have set up a Discord channel in a private server to receive the notifications. I create a webhook for this channel and store the webhook in Google Secret Manager. The GCF has access to this and sends the diary image to the channel. For example:
If I want to train a model to read my messy handwriting, I need to provide some labelled training data. To begin with, I will start by producing a mdel that classifies whole words from an image (as opposed to a more robust OCR model which might identify individual letters).
To help with this, I added a simple TKinter GUI that allows me to segment and label individual words from my scanned pages.
The resulting images are saved, for example:
Coming soon! Receiving the images daily is fine for now, but there is so much more I want to do with this project:
- Analysis of text. What words, phrases do I tend to use? Can I run a sentiment analysis or find trends?
- Train a NLP model on the text. Can I create a model that talks like me?