CVBABA

A multilingual résumé and cover-letter generation system with structured document output and ATS-aware workflows. Users describe their experience in natural language, choose an output language and template, and receive a formatted document.

Problem

Producing a résumé that is both correctly formatted and appropriate to a target language is tedious, and general-purpose models handle the layout half badly.

My contribution

A multi-step generation workflow separating content extraction from layout, so document structure stays stable across output languages and templates.

Role: Founding AI/ML engineer — team project.

Method

Input is parsed into a structured profile, adapted to the selected output language and target role, and rendered through templates that control layout independently of content. Options include ATS-oriented formatting and an optional photo. A live PDF preview shows the rendered document as settings change.

Evaluation

No formal evaluation was conducted. Output quality across languages, factual fidelity to user input, and ATS compatibility were checked informally during development and were never measured systematically.

Outcomes

• Working generation pipeline across multiple output languages and templates
• Live PDF preview with ATS-oriented formatting options
• Entrepreneurship Award from the Force-N programme

Limitations

Generated content can drift from what the user actually supplied, which is the hard problem in this domain and was not solved. ATS behaviour varies by employer system and was never verified against real parsers. Output quality was uneven across languages.

Fluency is easy and fidelity is not. A résumé that reads well but overstates the user's experience is a defect, and detecting that automatically proved harder than any part of the layout work.