Edify AI-Google Maps for Your Career

Personalized AI career mentor analyzing their skills, building dynamic paths, optimizing resumes via a multi-agent pipeline, and conducting real-time interviews, all powered by Vertex AI.

Comment

Edify AI — Google Maps for Your Career

Inspiration

Most students don't fail because they lack talent. They fail because they don't know the route. Every year, 40% of engineering graduates are considered unemployable by industry — not because they can't code, but because they never knew what to learn, how to present it, or how to talk about it. The problem isn't talent. It's the absence of a guide. Google Maps doesn't make you a better driver. It gives you the route. We wanted to build that — but for careers. A system that knows where you are, where you want to go, and reroutes you in real time when things change.

What It Does

Edify-AI is a multi-agent AI career mentor that guides students from confusion to job-readiness through five core capabilities:

1. AI Video Teacher (Core Feature) The heart of Edify-AI is a conversational AI agent you can actually talk to — face to face, like a real tutor on a video call. It's not a chatbot. It's a live video agent powered by Gemini that understands your personal context: your skill level, your roadmap, your past sessions. You can ask it to explain a concept, challenge you with questions, or guide you through what to do next — and it responds conversationally, remembering everything about you. It reroutes the lesson in real time based on your replies.

2. AI-Generated Course Creation For every step in your roadmap, Edify-AI assembles a complete, structured course from scratch — curated YouTube content with 85% relevancy scoring, chapter-by-chapter outlines, interactive quizzes, a code editor for hands-on practice, and discussion forums. No generic catalog browsing. Only what's needed for this skill, at this level, right now.

3. Personalized Career Roadmap Upload your resume, LinkedIn, and target role. Gemini analyzes your skill gaps against live job market data and generates a step-by-step roadmap — not based on trends, but built specifically around your strengths and weaknesses.

4. Multi-Agent ATS Resume Builder Five specialized agents (Skills, Education, Experience, Projects, Certifications) run in parallel to analyze a job description and produce a tailored, ATS-optimized resume with a match score and targeted improvement suggestions.

5. AI Mock Interview Coach Conducts realistic interviews personalized to your resume and target role. Gemini evaluates not just your answers but your voice tone, sentiment, and body language via webcam — then gives structured, actionable feedback.

How We Built It

Built entirely on Google's AI stack, aligned with the Agentic AI track:

  • Gemini 2.5 Flash — core reasoning engine across all agents (roadmap generation, course creation, resume analysis, interview feedback, and the live video teacher)
  • Vertex AI Agent Builder — orchestrates the multi-agent system
  • MCP (Model Context Protocol) — acts as the shared brain: maintains student context, long-term memory, and semantic state across all agents so the video teacher and every other feature feel like one unified mentor
  • LangGraph + LangChain — agent orchestration runtime; agents use tool-calling to take real actions (web scraping, resume export, course assembly)
  • Qdrant — vector search engine powering the RAG pipeline for semantic course recommendations and context-aware responses
  • Tavus — powers the real-time conversational video agent (AI Video Teacher)
  • Google Cloud Run — containerized deployment with auto-scaling
  • Next.js 14 + tRPC — frontend and type-safe API layer
  • PostgreSQL + Redis — persistence and session caching

The flow: User → Next.js → tRPC API → LangGraph Orchestrator → Specialized Agents → Gemini 2.5 Flash → Results streamed back to UI.

Challenges We Ran Into

Making the video teacher truly personal — getting the video agent to carry full student context (goals, progress, weak areas) into every conversation, not just answer generic questions, required deep integration between MCP memory and the video agent runtime.

Agent coordination without losing context was the hardest infrastructure problem. Six agents need to share memory and not contradict each other. MCP solved this — but wiring shared semantic memory across agents in real time took significant iteration.

Resume generation at production quality — getting five parallel agents to produce a coherent, well-formatted resume without hallucinating experience or skills required strict output schemas and multi-step validation.

Accomplishments That We're Proud Of

  • A conversational AI video teacher that knows your personal learning context and holds a real back-and-forth dialogue — not a chatbot, an actual video agent
  • MCP-powered mentor memory — the AI remembers your goals, past sessions, and progress across every feature, making every interaction feel connected
  • End-to-end agentic pipeline from profile input to job-ready output, with no manual steps required
  • Fully deployed on Google Cloud Run with real users testing it

What We Learned

  • MCP is genuinely powerful for stateful, multi-agent applications — but designing the memory schema carefully upfront saves weeks of refactoring
  • Gemini 2.5 Flash's multimodal capabilities (audio + video + text in one call) are underutilized — combining them for both the video teacher and interview coaching unlocked quality we didn't expect
  • Agentic systems need graceful failure logic — agents must know when to ask for clarification rather than hallucinate an action

What's Next for Edify AI

  • AR/VR Classroom — peer collaboration in a virtual campus for immersive, hands-on learning
  • Institutional Dashboard — colleges and placement cells get aggregate analytics on student readiness and skill gaps
  • Live Opportunity Matching — agents that monitor job boards and hackathons in real time and notify students the moment they're ready for a specific role
  • Global Expansion — roadmaps calibrated to job markets across 30+ countries

Built With

Share this project:

Updates