SkillBridge AI

Students blocked by ATS,no feedback on skill gaps.Unprepared for interviews, low placement rates.SkillBridge AI delivers personalized, data‑driven career prep and institutional insights.

Comment

Inspiration

The frustration of watching talented people get rejected by automated systems before a human ever sees their resume drove us to build SkillBridge AI. 75% of resumes never reach a recruiter due to rigid ATS filters. We wanted to level the playing field — giving every student the same quality of career coaching that was previously only available to a privileged few.

What it does

SkillBridge AI is an agentic career platform that guides students from resume to placement:

  • Resume Analyzer — ATS score out of 100, keyword gaps, improvement suggestions
  • JD Matcher — match percentage and skill gap breakdown against any job description
  • Learning Roadmap — personalized week-by-week plan across 4, 8, or 12 weeks
  • Adaptive Quiz — domain-specific questions that scale in difficulty in real time
  • Mock Interviews — Technical or HR mode with scored, structured AI feedback
  • Job Search — AI-matched recommendations ranked by resume compatibility
  • Admin Dashboard — cohort analytics, eligibility control, and executive summaries

How we built it

Frontend: React 18 + Vite, Framer Motion, Recharts, Axios

Backend: FastAPI + Uvicorn, Pydantic, PyMuPDF, scikit-learn, Redis

Database: Supabase PostgreSQL with JSONB and Row Level Security

AI Inference: Oxlo APIs powering a custom AI Model Router

Model Task
DeepSeek R1 8B Resume & ATS parsing
DeepSeek V3.2 JD matching
GPT-OSS 20B Roadmap generation
DeepSeek Coder 33B Quiz generation
Gemma 3 27B Mock interviews

Challenges we ran into

  • Model routing logic — deciding which model to call for each task and structuring prompts to return consistent, parseable JSON across five different models
  • ATS accuracy — making the scoring feel meaningful and actionable rather than arbitrary
  • Adaptive difficulty — calibrating quiz difficulty in real time without making jumps feel jarring to the student
  • Admin scale — designing a dashboard that handles large student cohorts without performance degradation
  • Prompt engineering — getting the mock interview model to evaluate nuanced responses the way a real interviewer would

Accomplishments that we're proud of

  • Built a fully working end-to-end career platform in a single hackathon sprint
  • Designed and implemented a custom AI Model Router that intelligently routes five different Oxlo-powered models
  • Achieved consistent structured JSON output across all five models for seamless frontend rendering
  • Built a real-time adaptive quiz engine that adjusts difficulty question by question
  • Delivered an admin dashboard with live cohort analytics, eligibility filtering, and AI-generated executive summaries

What we learned

  • Different LLMs have very different strengths — routing by task rather than using one model everywhere makes a measurable difference in output quality
  • Prompt structure matters as much as model choice — small changes in framing dramatically affect consistency and accuracy
  • Building for institutions requires a completely different UX mindset than building for individual users
  • Real-time evaluation of open-ended responses is genuinely hard — scoring interviews fairly required far more iteration than expected

What's next for SkillBridge AI

  • Voice interviews — real-time spoken mock interviews with live transcription and evaluation
  • LinkedIn & GitHub integration — auto-enriching student profiles with live data
  • Company-specific prep tracks — tailored interview preparation for target companies
  • Mobile app — React Native version for on-the-go learning and practice
  • Peer mock interviews — scheduling system for student-to-student practice sessions
  • Salary negotiation simulator — AI-powered practice for offer negotiation conversations
  • ML-powered placement predictions — forecasting a student's placement probability based on their activity and scores

Built With

Share this project:

Updates