Inspiration
Refine was inspired by a common problem students and early-career applicants face: getting rejected without knowing why. A resume may look polished, but it can still miss the exact evidence, keywords, level-fit signals, or project details a job description expects. We wanted to build a tool that gives applicants a clearer, calmer way to understand the gap before they apply again.
What it does
Refine analyzes a resume against a target job description and turns the result into an application improvement workflow. It gives a fit diagnosis, ATS readiness checks, evidence gaps, resume patch suggestions, a weekly action roadmap, job description comparison, and adaptive interview practice. The goal is not just to produce a score, but to explain what to fix, why it matters, and what to do next.
How we built it
We built Refine as a FastAPI and Jinja web app powered by Google ADK and Gemini. The backend parses resumes and job descriptions, separates hard requirements from softer coaching signals, maps resume evidence to role expectations, and uses guarded Gemini reviews to improve scoring and feedback. We also added progress-based diagnosis, resume preview parsing, Google login, saved history, Cloud Run deployment, and a polished UI with light/dark mode and custom branding.
Challenges we ran into
One major challenge was making the scoring feel fair and useful. Early versions penalized generic phrases like “ability to manage tasks” too heavily, so we rebuilt the scoring system to focus on hard skills, real evidence, level fit, and guarded recommendations. Another challenge was interview feedback: the first version surfaced raw resume fragments like contact details, so we added sanitization and structured coaching sections. We also had to improve performance by splitting diagnosis into faster staged results.
Accomplishments that we're proud of
We are proud that Refine became more than a basic resume checker. It now guides users through a full application workflow: diagnose, patch, plan, compare roles, export, and practice interviews. We are also proud of the adaptive interview simulator, the more genuine scoring system, and the clean user experience that helps applicants know what to do next instead of overwhelming them with options.
What we learned
We learned that useful AI career tools need strong guardrails. A model can generate helpful feedback, but deterministic checks, evidence validation, fallback logic, and privacy-aware filtering are essential. We also learned how important workflow design is: users do not just need information, they need the right next step at the right time.
What's next for Refine
Next, we want to make Refine even more personalized. Future improvements could include deeper portfolio analysis, richer project evidence suggestions, recruiter-style mock interviews, better analytics across saved applications, and more collaborative tools for mentors or career coaches.
Log in or sign up for Devpost to join the conversation.