Now●

A dated log of what I built, what broke, and what I finally understood. Written per milestone, unedited afterwards - the wrong turns are the useful part.

Building

Xylodesk — Resume Studio's last stretch: Tailor with AI, resume upload, and cover letters.

Learning

React, Rails, OOP, and system design; mostly by building Xylodesk, without much of a structured path.

Next up

Tailor with AI, for rewriting a resume against a specific job posting, and resume upload, so nobody starts from scratch or re-enters their data twice.


week 35

Five templates shipped, and a cache that did not know they existed

Registered five named resume templates and retired the old generic one, to clear a hard requirement: at least five templates rated ATS-friendly. Building them surfaced a real bug: applying a new template did not invalidate the public share-link's cached PDF, so a candidate switching templates could keep serving their old design to anyone with the link, indefinitely.

Resume StudioCaching

week 35

A PDF that was 3.5% smaller than it should have been, silently

A banner résumé template pulled its header toward the page edge with a negative margin — looked right in the browser canvas. In the exported PDF, Chromium's print engine clipped the bled content entirely and silently shrank the whole document to fit, with no error. Fixed by shrinking the page's own padding instead of fighting it with a negative margin.

Resume StudioPDF export

week 33

AI resume suggestions, narrowed on purpose

The suggestions/rewrite feature only round-trips on text that already exists in the resume, not free-form generation. Contact fields (email, phone, name) are excluded from what ever reaches the prompt, even if someone tried to pass one in.

Resume StudioGemini

week 33

Message requests: two inbox tabs, not three

The written design put pending message requests in their own tab, mirroring the Instagram/LinkedIn pattern. Once built, that added friction without a real benefit — a request is still a DM, just unanswered. Shipped two tabs instead of three, with the pending/active split staying real on the backend even though the frontend merges the rows.

MessagingProduct

week 29

Heuristic matching first, AI second

Shipped skills and experience overlap scoring before touching Gemini. The AI tier now layers on top and falls back to the heuristic when a call fails. A working, explainable baseline beats a clever thing that sometimes returns nothing.

MatchingGemini

week 28

A <label> around a dropdown breaks focus

Several profile-editor dropdowns kept jumping back to an earlier field after picking a value. Root cause: a <label> is meant for exactly one control, but these components render extra buttons inside once opened, the browser was forwarding focus to the first one it found. Fixed with a plain wrapper for any field that isn't a bare input.

ProfileFrontend

week 23

Devise was vestigial; removed

A routine dependency bump turned up that Devise was never actually wired into authentication — auth had been a custom JWT service, has_secure_password, the whole time. Removed the gem, its orphaned transitive deps, and a stale code comment rather than upgrade something with zero runtime references.

AuthenticationCleanup

week 22

Made tenant scoping fail loudly instead of quietly

Switched acts_as_tenant to fail-closed. An unscoped query now raises instead of returning an empty set — because an empty set looks like "no results" in the UI and like a passing test in CI.

RailsMulti-tenancy

week 22

Wrote a reviewer instead of writing more code

Spent the day building subagents that review my diffs rather than shipping features. Felt slow. It has since caught authorization gaps and tenant-isolation issues I personally wrote and personally approved.

ToolingReview

week 22

GDPR erasure is not a DELETE statement

Candidates can erase their data; employers must retain some of it. Landed on a 30-day window that anonymizes retained records in place, so foreign keys survive and the person does not. The hardest part was accepting that the naive answer was wrong.

ComplianceData

week 22

Job postings and applying, shipped the same day

Job posting CRUD for recruiting orgs and the apply flow for candidates landed together — a posting needs a lifecycle (draft/published/closed) and a recruiting-staff-only write path before a single candidate ever sees it. The one rule that had to be right from day one: "one application per job per candidate" is a DB unique index, not a Rails validation — a validator alone races under concurrent submits.

Job BoardApplications

week 20

Split the two things I kept calling a "profile"

A candidate profile and an employee HR record looked like one model for about a week. They are not: one is public, one holds compensation. Splitting them early cost an afternoon and removed an entire category of future leak.

SchemaPrivacy