Edify AI-Google Maps for Your Career

The shortest path from 'Student' to 'Hired'. Career paths aren't linear, but they shouldn't be a maze. Edify-AI is your personal career navigator-an AI mentor that maps your growth to success.

Comment

Edify-AI

Google Maps for Your Career

Students don't fail because they lack talent - they fail because they don't know the route.

Edify-AI is an AI-powered career advisor that guides students from confusion to clarity - a personalized route, real-time guidance, and rerouting when they go off track.

Inspiration

We watched friends struggle through their career journeys - jumping between random tutorials, investing in courses that didn't match their level, building generic resumes that got rejected, and failing interviews despite knowing the material.

The problem wasn't effort. It was lack of direction.

Every existing platform solves one piece of the puzzle. Udemy sells courses. LinkedIn helps with networking. ChatGPT gives answers. But no one connects the entire journey from "I'm confused" to "I got the job."

We asked: What if career guidance worked like Google Maps? - showing you where you are, where you want to go, the best route, and rerouting when you take a wrong turn.

What It Does

Edify-AI provides a complete 6-step career journey:

1. Personalized Career Roadmap
Analyzes your resume, skills, and goals. Pulls context from GitHub, LinkedIn, YouTube, Roadmap.sh. Generates a personalized learning path - not generic advice, but a route built for you.

2. AI-Generated Learning Modules
Custom courses for each roadmap step with curated video content, practice environments, and quizzes matched to your current level.

3. Job-Specific Resume Builder
Parses job descriptions and generates ATS-optimized resumes - automatically updated as your skills improve.

4. AI Mock Interviews
Realistic simulations that analyze both your answers AND body language/confidence via video. Not ready? We reroute instead of reject.

5. Real-Time AI Video Mentor
A video mentor - not a chatbot. It understands your context, remembers your progress, and guides you through personalized interaction. When you're stuck, you ask.

6. AR-Based Classrooms
Immersive 3D learning spaces for virtual study groups and peer collaboration. Learning becomes social, not isolated.

At the core: MCP (Model Context Protocol) - our custom protocol connecting everything into one intelligent mentor that adapts as you progress.

How We Built It

Multi-Agent GenAI Architecture

We built 6 specialized LangChain agents orchestrated by a central coordinator:

Agent Function Technique
Orchestrator Agent Routes requests, manages state, coordinates all agents MCP Protocol + state machine
Roadmap Agent Skills + market data → personalized learning path Chain-of-thought reasoning
Course Agent Generates custom learning modules from curated content RAG + content synthesis
Resume Agent JD parsing → ATS-optimized resume generation Few-shot examples
Interview Agent Dynamic Q&A + real-time evaluation Context-aware prompting
Mentor Agent Conversation memory + personalized guidance Self-consistency checks 
User Request → Orchestrator → Routes to Specialist Agent(s) → Aggregates Response → User
                    ↓
              Maintains context across entire journey via MCP

Guardrails on every call: Input validation, hallucination detection, fallback responses, rate limiting.

Video Intelligence

Model Purpose
YOLO Real-time body language, eye contact, confidence signals
Tavus Conversational video AI - the mentor that sees and responds to you

Infrastructure

Component Choice Why
API tRPC End-to-end type safety
Background Jobs Inngest Non-blocking AI workloads
Agent Runtime Python + LangChain Best AI ecosystem
Frontend Next.js SSR, React, fast
Database PostgreSQL ACID + JSON flexibility
Infra GCP Cloud Run Auto-scaling, pay-per-request
CI/CD GitHub Actions + Cloud Build Automated deployment
Scraping Apify GitHub, LinkedIn, YouTube data

DevOps Pipeline

Push → GitHub Actions → Tests → Docker Build → GCP Cloud Run → Health Check → Live

🕷️ The Data Engine: Powered by Apify

Most AI career advisors rely on static, outdated LLM training data. Edify-AI is different. We use Apify as our "Live Web Crawler" to ensure every learning path, resume, and interview is based on the market today, not two years ago.

How We Leverage Apify:

Actor Used Purpose Impact
LinkedIn Scraper Pulls real-time Job Descriptions (JDs) for our Resume Agent. Matches your resume to actual, active job requirements.
GitHub Scraper Analyzes user repositories and contribution history. Provides objective, data-backed skill assessments for the Roadmap.
YouTube Scraper Curates the highest-rated teaching materials. Ensures Course Agents use the most relevant, up-to-date video tutorials.
Google Search Tracks emerging trends in tech and the job market. Allows Edify-AI to "reroute" you if a skill becomes obsolete.

Bridging the "Context Gap"

By integrating Apify directly into our MCP (Model Context Protocol), our agents don't just "guess" what a Front-end Engineer needs. They crawl LinkedIn, analyze the most common requirements for current openings, and dynamically adjust your learning path.

Apify isn't just a scraper for us—it's the sensory system that keeps Edify-AI grounded in reality.

Impact

Feature Traditional Edify AI Time Saved
Resume Analysis 45-60 min 2-3 min 95%
Learning Path Research 1-2 hours 3-5 min 96%
Interview Preparation 1-2 hours 3-5 min 95%
Skill Gap Identification 30-45 min 1-2 min 96%
Course Generation 1-2 hours 2-4 min 97%
Total 5-10 hours ~20 min ~96%
Stakeholder Value
Students Democratizes career guidance, real interview readiness
Institutions Scalable placement infra, 60-70% less counselor load
Society Reduces unemployment, promotes equity

Challenges We Ran Into

1. Multi-Agent Coordination
Getting 4 specialized agents to share context without conflicts was hard. We solved it with MCP - a custom protocol that maintains state across all agents.

2. Video Analysis Latency
YOLO inference was too slow for real-time feedback. We moved to async processing via Inngest, showing results after the interview instead of during.

3. Hallucination in Career Advice
Early versions gave confident but wrong career suggestions. We added guardrails: few-shot examples, self-consistency checks, and confidence thresholds that trigger fallback responses.

4. Scraping Rate Limits
LinkedIn and GitHub have aggressive rate limiting. We implemented intelligent caching and batch processing through Apify to stay within limits while keeping data fresh.

5. AR Performance on Mobile
AR.js was heavy on mobile browsers. We optimized by lazy-loading 3D assets and reducing polygon counts for classroom environments.

Accomplishments We're Proud Of

  • End-to-end working prototype - not mockups, real AI integrations
  • 4 specialized LangChain agents working together seamlessly via MCP
  • YOLO-based body language analysis in mock interviews
  • Production deployment on GCP with full CI/CD pipeline
  • AR classrooms running in browser without native apps
  • Guardrails that actually work - zero hallucination incidents in testing
  • Built the entire system in a hackathon timeframe

What We Learned

On AI Architecture: Single LLM calls don't scale for complex workflows. Multi-agent systems with specialized roles produce better results than one "do everything" prompt.

On Prompting: Few-shot examples + chain-of-thought + self-consistency checks = reliable outputs. Guardrails aren't optional — they're essential.

On Infrastructure: Inngest is incredible for AI workloads. tRPC eliminates an entire class of bugs. GCP Cloud Run handles scale without DevOps overhead.

On Product: Users don't want more tools - they want fewer tools that work together. The "one mentor" mental model resonates because it mirrors how real career guidance works.

What's Next for Edify-AI

Short-term:

  • Voice-based mentor interactions (currently text + video)
  • Integration with actual job boards for real-time market data
  • Mobile app for on-the-go learning

Medium-term:

  • B2B pilot with 2-3 colleges for placement infrastructure
  • Expanding AR classrooms with more interactive elements
  • Adding peer-to-peer mentorship matching

Long-term:

  • White-label solution for educational institutions
  • API for third-party integrations
  • Expanding beyond tech careers to other fields

Feasibility & Scalability

Why This Works at Scale

Aspect Our Approach Scalability
Compute GCP Cloud Run Auto-scales 0→∞, pay-per-request
AI Workloads Inngest queues Async processing, no blocking
Database PostgreSQL + Redis Horizontal scaling ready
Agents Stateless Python Spin up unlimited workers
Cost ~$0.08/user/month Sustainable unit economics

Production-Ready Today

  • Docker containerized - consistent environments everywhere
  • CI/CD pipeline - automated testing + zero-downtime deploys
  • Health checks - automatic rollback on failures
  • Rate limiting - protects against abuse and cost spikes

Market Opportunity

Metric Value
Global TAM 200M+ college students seeking career guidance
EdTech Market $50B+, growing 15% YoY
Career Services $10B spent annually by institutions
Pain Point 70% students report career anxiety as top concern

Business Model

Segment Model Revenue
B2C (Students) Freemium → $9.99/mo premium High volume, low touch
B2B (Colleges) $5-15/student/year site license Recurring, high retention
Enterprise White-label API licensing High margin partnerships

Why Now?

  • GenAI maturity - LLMs finally good enough for personalized guidance
  • Post-COVID shift - students comfortable with digital-first learning
  • Hiring crisis - employers struggling to find job-ready graduates
  • Counselor shortage - 1 counselor per 400+ students in most colleges

Technical Stack

Layer Technology
Frontend Next.js, React, AR.js
API tRPC (type-safe)
Backend Node.js, Python (LangChain)
AI/ML OpenAI, LangChain, YOLO, Tavus
Jobs Inngest
Scraping Apify
Infra GCP, Docker, CI/CD
Protocol Custom MCP

We don't sell courses. We don't sell tools. We give direction.

Edify-AI is Google Maps for your career.

Built With

Share this project:

Updates