Sage | Devpost

Inspiration Hiring teams are overwhelmed by application volume and increasingly polished AI-generated resumes. At the same time, candidates often struggle to communicate the real decisions, ownership, and impact behind their work through a resume or cover letter alone. We built Sage to make the process more human and evidence-based in both directions: recruiters get a faster way to navigate application volume, while candidates get a meaningful opportunity to advocate for their experience. What it does Sage conducts a structured, adaptive interview based on a candidate’s application claims. It asks focused questions, supports text and audio answers, preserves transcript corrections, and allows candidates to attach supporting evidence. The system produces a claim-by-claim evidence report showing:

What the candidate claimed
What Sage asked
How the candidate responded
What evidence was attached
What remains demonstrated, partially demonstrated, or unresolved
Limitations that recruiters should consider Recruiters can review multiple candidates, filter by role and status, inspect evidence timelines, view execution traces, and record their own next step. Sage does not assign hiring scores or make hiring recommendations. How we built it We built the candidate experience with Next.js, React, and scoped Sage styling. The recruiter experience uses the same frontend application with a review queue, report views, evidence timelines, claim detail modals, source toggles, print views, dark mode, and recruiter-controlled next steps. The backend is built with FastAPI and provides application, interview, answer, evidence, report, and trace endpoints. Demo mode uses deterministic data so the entire flow can run reliably without external model keys or third-party services. We added browser and backend tests covering the full journey from creating an application through completing an interview, attaching evidence, generating a report, and handing that report to the recruiter. Challenges we ran into The biggest challenge was integrating multiple product surfaces into one coherent flow. The candidate interview, recruiter queue, backend report, evidence system, and execution trace each started with different assumptions about IDs, data shape, and navigation. We also had to preserve privacy boundaries around audio and camera use. Camera preview remains local and optional, audio is used for transcription, and no facial, vocal, emotional, or behavioral analysis is performed. Another challenge was making the demo deterministic while still showing a realistic interview. We needed the system to support live API behavior, controlled fixtures, offline rehearsal, transcript correction, evidence limitations, and graceful backend failure. Accomplishments that we're proud of We are proud that Sage creates a complete, inspectable loop from candidate claim to recruiter decision support. Candidates can explain their work in their own words, correct transcripts, attach evidence, and see exactly what will be handed to the recruiter. Recruiters can then move from a queue to a claim-level report without losing the original context. We are also proud of the transparency built into the product. Reports preserve limitations, separate evidence from assessment, retain original transcripts, expose session traces, and avoid opaque scores or automated hiring recommendations. Finally, we built a working end-to-end demo that can be run locally and tested repeatedly with deterministic fixtures. What we learned We learned that trustworthy hiring tools need to show their work. A conclusion is much more useful when the recruiter can trace it back to the candidate’s claim, question, answer, and evidence. We also learned that candidate advocacy and recruiter efficiency are not opposing goals. A structured interview can reduce recruiter workload while giving candidates more room to communicate context that resumes leave out. Most importantly, we learned that careful product boundaries matter. The system becomes more trustworthy when it clearly states what it observes, what it does not analyze, what evidence is limited, and where human judgment is still required. What's next for Sage Next, we want to connect Sage to real applicant tracking systems and support authenticated recruiter workspaces. We would also like to add configurable interview templates, stronger evidence provenance, candidate-controlled sharing, and persistent recruiter decisions with an audit trail. We plan to improve accessibility, multilingual support, and accommodations for candidates who prefer text, audio, or other communication formats. Longer term, Sage could help teams compare evidence across roles and interviews while preserving human review, candidate context, and clear limitations. The goal is to make hiring more efficient without making it less human.