Repository navigation
feat: read older job emails to fill in past applications, and never take a job alert for one - #77
Merged
Merged
Conversation
…ake a job alert for one
The Inbox can read 6 months, 1 year or 2 years back. It shows the count first, runs in the background with progress and
a stop button, and carries on where it stopped. An application is added only when an email shows it was made, marked
as from email, with its path rebuilt from its emails: being received or turned down is enough, while an interview, test
or offer also needs an applicant tracking system or a company already shown. Old emails fill in an application's history
without moving it back, and ask nothing of the user.
Job alerts and recommendations are now recognised by sender and subject, in the regular sync too, and never reach the
model: "Junior Application Software Engineer at X: up to £45K/year" matched only on its title. Real confirmations from the
same senders ("your application was sent to X") still count.
…wn interview invitation counts The model took Indeed for the employer, or filled blanks with "Unknown". A job board's confirmation naming only the role now goes under "Employer not named (via Indeed)", and an email naming no role adds the application as "Role not stated" unless another of its emails names it. An interview from the employer's own address shows they applied, as a recruitment agency's first approach doesn't.
The Applications page gains a third view, How they went: from Applied through Responded and Interview to Offer, Rejected, No reply (nothing for 4 weeks), Still waiting or Closed by you, over 3 months to 2 years. It takes each application's path from its status history, and it shows how many heard back. Hovering or focusing a flow gives its count, and the figures are also there as a table. It says what it can't know: applications with no confirmation email, and steps by phone.
feat: a Sankey chart of how applications went
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Why
Windrush has only read job emails since Gmail was connected, so earlier applications are missing. Application-style subjects also catch job alerts: "Junior Application Software Engineer at X: up to £45K/year" matches only because of its title. With past applications added automatically, an alert must never become one.
What changed
The backfill search leaves out tracked company names, which mostly matched mail that isn't about jobs: 311 emails for 6 months, against 1,231 with them.
Results
A read-only trial on real mail ran the alert filter over the 156 unread emails from the last 6 months and the reader on a random 60 of the rest. No alert would be added: the filter caught 6, and the model rejected 14 more unrelated emails. The trial found the name problems fixed above, and the re-run showed them fixed.
The real-model eval, now 20 emails including new alert cases, scored 20/20 and then 19/20 after the final prompt change. The miss read a "your application was sent to" sample as unrelated; the real ones in the trial were read correctly.
Testing
Merge order
Second of three. Based on
feat/status-history.