Inspiration# About Prompt2Resume
When I first entered college, I had no idea how to build a well-formatted, concise, and professional resume. I spent hours experimenting with different layouts, trying to make my resume presentable. I created Prompt2Resume so that other students wouldn’t face the same problem I did; it helps match a user’s resume to the format of some of the best resumes in the industry. The tool supports multiple styles of technical resumes, with different sections and layouts, making it flexible for a variety of use cases.
Inspiration
The inspiration came from my own struggle: entering college without guidance on how to structure a resume effectively. I wanted to build something that could save students hours of manual formatting and help them present their best work in a clean, industry-standard layout.
What I Learned
This project taught me a lot about technologies I had never used before:
- OCR (Optical Character Recognition): I learned how to extract text and coordinates from PDFs, which enabled me to accurately map sections from a user’s resume, although I ended up not using this.
- NLP (Natural Language Processing): I integrated NLP to automatically map sections from the uploaded resume to the template, deciding the best placement for each piece of information. This taught me how to combine heuristics with machine learning models for real-world text processing.
- End-to-End Pipeline Development: Prompt2Resume was my first complete pipeline project for a hackathon. I learned how to structure data flow, manage intermediate metadata, and produce a final PDF output.
Through this project, I also gained hands-on experience in debugging, iterative design, and workflow automation.
How I Built It
I started building the project with vibe coding using ChatGPT, which allowed me to experiment with features quickly. When Kiro became available, I switched to using it for generating specs, design documents, and tasks. This approach let me watch features get integrated automatically, while still keeping control over the high-level design.
I structured conversations with Kiro carefully:
- Trial-and-error initially — asking Kiro to implement features as I went, which was fun but inefficient.
- Spec-driven approach — defining designs, requirements, and tasks upfront. I double-checked with ChatGPT to ensure nothing was missing. This improved efficiency and helped avoid inconsistencies.
The most impressive Kiro-generated code included coordinate mapping for OCR and NLP integration for section mapping, which were critical for automating template-to-resume conversion.
Challenges and Frustrations
- Inconsistencies with AI: Both vibe coding and Kiro were sometimes unpredictable. Kiro would create new pipelines instead of merging with existing code, leading to duplicate files and messy structure.
- Credit limits: Running Kiro sometimes exceeded usage limits unexpectedly.
- Double-checking goals: I often had to manually verify that the AI-generated implementation actually achieved what I intended, rather than just following the spec.
Despite these challenges, the process taught me patience, careful planning, and how to structure AI-assisted development effectively.
Outcome
Prompt2Resume now automates the process of formatting resumes according to chosen templates, saving students hours of manual work and giving them a polished, professional output. It’s a first step toward creating a fully editable, version-controlled resume builder that integrates OCR, NLP, and customizable templates, all of which are lessons I learned while tackling the project from scratch.
Accomplishments that we're proud of
I am proud to have accomplished such a big feat — learning new technologies and interests such as NLP, exploring OCR, and seeing firsthand how far vibe coding can really take a project. This experience not only strengthened my technical skills but also taught me how to structure AI-assisted development, troubleshoot inconsistencies, and iterate effectively on a complex end-to-end pipeline.
What's next for Prompt2Resume
Implement Version Control style journaling for generated pdfs, integrate with React and improve UI. Add stricter logic to make sure things fit on one page!
Log in or sign up for Devpost to join the conversation.