Taylor.io

Tailoring a resume can help, but the normal process is slow and scattered. The user reads a job description in one place, asks an AI tool questions in another, edits a document somewhere else, and repeats the same background information every time.

Most resume products also treat the resume like one block of text. I wanted the underlying career information to be structured, reusable, and able to shift without forcing the user to rebuild everything from the beginning.

I wanted the product to feel guided instead of opening on an empty document. Users should be able to see the story they already have, understand what the system is emphasizing, and make smaller decisions instead of rewriting the entire resume in one large text box.

The editor and preview stay connected because the final document still matters. The structured data is useful behind the scenes, but the user needs to see how those decisions actually change the resume a recruiter will receive.

I also started giving the AI output a clearer explanation layer. A suggestion is easier to trust when the user can see what was prioritized, what was trimmed, and where the system was intentionally being cautious.

The core system is organized around structured resume data rather than a single generated document. Education, experience, projects, skills, and layout settings can be edited independently and reused across different versions.

FastAPI handles the parsing, tailoring, authentication, and export workflows. PostgreSQL stores user and resume data, while React powers the editor and preview experience.

The AI layer handles extraction, analysis, and drafting, but the output is constrained by the user's existing evidence so the product does not casually invent experience or replace the user's real stack with keywords from a job description.

Playwright and a separate DOCX builder handle the two export paths because a good browser preview does not automatically translate into a reliable editable document.

The biggest product challenge was my own scope creep. Every useful idea made me want to add three more before I had proven the main flow.

The hardest system problem was balancing useful rewriting with strict truth. The product has to make a resume more focused without turning it into a confident list of things the person never actually did.

Document generation was another project by itself. A resume can look correct in the browser and still break once it becomes a PDF or DOCX file with different page, spacing, and font behavior.

Taylor.io is now an early working product instead of an idea I keep redesigning privately. It can take existing resume information, structure it, tailor the presentation for a role, and export a usable document.

There are still imperfections, but shipping it turned a private redesign loop into a real system that can be tested, evaluated, and improved. It also became the clearest example of how I approach a product from the first frustration through the data model, interface, AI behavior, and deployment.

The next version needs a cleaner onboarding flow, stronger version comparison, and an easier way to move between a base profile and several tailored resumes.

I also want to keep improving the explanation and evaluation layers so users can understand why a change was made, catch weak suggestions quickly, and stay in control of the final story.