⚡ SkillBridge — AI-Adaptive Onboarding Engine

ARTPARK CodeForge Hackathon Submission
An AI-driven system that analyzes a candidate's resume against a job description, identifies precise skill gaps, and generates a personalized, prioritized learning roadmap.

---

🎯 Problem Statement

Corporate onboarding suffers from a one-size-fits-all problem:

• Experienced hires waste time on concepts they already know
• Beginners get overwhelmed by advanced modules
• No one measures the actual gap between hire capabilities and role requirements

SkillBridge solves this by generating a personalized learning pathway in under 100 seconds using AI.

---

🚀 Live Demo

Upload a resume (PDF/DOCX) + job description (PDF/TXT) → get a fully personalized roadmap with reasoning traces.

---

✨ Features

Feature	Description
🤖 Intelligent Parsing	Hybrid LLM + NER + alias extraction from resume & JD
📊 Skill Gap Analysis	3-layer comparison: exact match → skill family → cosine similarity
🗺️ Adaptive Pathway	Original WGT (Weighted Graph Traversal) algorithm
🔍 Reasoning Traces	Every recommendation explained with evidence from your resume
🎨 Interactive Roadmap	react-flow DAG with color-coded nodes and filter controls
🚫 Zero Hallucinations	All course recommendations from curated closed catalog only
⚡ Real-time Progress	Live 6-stage processing pipeline with WebSocket-style polling

---

🏗️ Architecture

┌─────────────────────────────────────────────────────────┐
│              FRONTEND (React 18 + Vite)                 │
│   Upload → Processing → Dashboard → Roadmap (DAG)       │
└──────────────────────────┬──────────────────────────────┘
                           │ REST API
┌──────────────────────────▼──────────────────────────────┐
│               BACKEND (FastAPI + Python 3.11)           │
│  /analyze  /status  /results  /trace  /health           │
└──────┬──────────────────────────────────────────────────┘
       │
┌──────▼──────────────────────────────────────────────────┐
│                    AI / ML LAYER                        │
│  Parser (LLM + spaCy)  →  Gap Analyzer (FAISS)         │
│  WGT Algorithm (NetworkX)  →  Trace Generator (LLM)    │
└─────────────────────────────────────────────────────────┘

---

🧠 Skill Gap Analysis Logic

3-Layer Matching Pipeline

For each JD skill requirement:

Layer 1 — Exact ID Match
  Resume has exact skill? → compute level coverage score
  coverage ≥ 0.85 → fully covered (skip)
  coverage < 0.85 → weak gap

Layer 2 — Skill Family Match  
  e.g. Resume has PostgreSQL + MySQL → covers SQL requirement (0.84 coverage)
  e.g. Resume has React → partial credit for Next.js
  Prevents false gaps from skill variant naming

Layer 3 — Semantic Similarity (cosine)
  Embed all skill names with all-MiniLM-L6-v2 (384-dim)
  FAISS IndexFlatIP similarity search
  score ≥ 0.62 → weak gap
  score < 0.62 → missing gap

WGT Priority Score Formula (Original Algorithm)

P(skill) = 0.40 × gap_severity
         + 0.30 × requirement_weight  
         + 0.20 × dependency_urgency
         + 0.10 × experience_penalty

Where:
  gap_severity       = 1.0 - coverage_score
  requirement_weight = 1.5 (required) or 1.0 (preferred), normalized
  dependency_urgency = blocked_gap_skills / total_gap_skills
  experience_penalty = reduced for trivial skills on senior hires

Skills are then ordered via Kahn's topological sort with P-score tie-breaking, ensuring prerequisites always appear before dependent skills.

---

🛠️ Tech Stack

Frontend

Library	Version	Purpose
React	18.3	UI framework
Vite	5.x	Build tool
@xyflow/react	11.x	DAG roadmap visualization
recharts	2.x	Skill gap charts
react-dropzone	14.x	File uploads
zustand	4.x	State management

Backend

Library	Version	Purpose
FastAPI	0.111	REST API framework
SQLAlchemy	2.0	ORM + SQLite storage
pdfplumber	0.11	PDF text extraction
python-docx	1.1	DOCX text extraction

AI / ML

Model / Library	Purpose
Mistral-7B-Instruct (Ollama) or GPT-4o-mini	Skill extraction + reasoning traces
all-MiniLM-L6-v2 (sentence-transformers)	384-dim skill embeddings
FAISS (faiss-cpu)	Vector similarity search
spaCy en_core_web_lg	NER validation
NetworkX	Skill dependency DAG

Datasets & Ontologies

Dataset	Source	Rows Processed	Usage
Kaggle Resume Dataset	snehaanbhawal	2,484 resumes	Enriched 11 skill aliases in taxonomy
Kaggle JD Dataset	kshitizregmi	2,277 JDs	Computed demand weights for WGT P-Score
O*NET Database 28.1	onetcenter.org	Full database	Canonical skill IDs + prerequisite graph

---

⚙️ Setup Instructions

Option A — Docker (Recommended for Judges)

# 1. Clone the repository
git clone https://github.com/your-team/skillbridge.git
cd skillbridge

# 2. Add your OpenAI API key
# Edit skillbridge/backend/.env.docker
# Set: OPENAI_API_KEY=sk-your-key-here

# 3. Start everything
docker compose up --build

# 4. Open the app
# Frontend: http://localhost:3000
# API Docs: http://localhost:8000/docs

Option B — Local Development

Prerequisites

• Python 3.11+
• Node.js 18+
• Ollama (for local LLM) OR OpenAI API key

Backend Setup

cd skillbridge/backend

# Create virtual environment
python -m venv venv
source venv/bin/activate        # Windows: venv\Scripts\activate

# Install dependencies
pip install -r requirements.txt

# Download spaCy model
python -m spacy download en_core_web_lg

# Configure environment
cp .env.example .env
# Edit .env — set LLM_PROVIDER and API key

# Start server
uvicorn app.main:app --reload --port 8000

Frontend Setup

cd skillbridge/frontend

# Install dependencies
npm install

# Configure environment
# Edit .env — set VITE_API_URL=http://localhost:8000

# Start dev server
npm run dev
# Opens at http://localhost:5173

LLM Setup (Local — Free)

# Install Ollama from https://ollama.ai
ollama pull mistral

# Update backend/.env:
# LLM_PROVIDER=ollama
# OLLAMA_BASE_URL=http://localhost:11434
# OLLAMA_MODEL=mistral

---

📁 Project Structure

skillbridge/
├── backend/
│   ├── app/
│   │   ├── api/
│   │   │   └── routes.py          # All API endpoints
│   │   ├── core/
│   │   │   ├── config.py          # Settings management
│   │   │   ├── database.py        # SQLite/PostgreSQL setup
│   │   │   └── llm.py             # Unified LLM caller
│   │   ├── data/
│   │   │   ├── skill_taxonomy.json    # 71 skills + prerequisites
│   │   │   └── course_catalog.json   # 58 curated modules
│   │   ├── models/
│   │   │   ├── db_models.py       # SQLAlchemy ORM models
│   │   │   └── schemas.py         # Pydantic schemas
│   │   ├── services/
│   │   │   ├── parser.py          # Resume + JD extraction
│   │   │   ├── embeddings.py      # Sentence transformer service
│   │   │   ├── gap_analyzer.py    # 3-layer gap analysis
│   │   │   ├── wgt_engine.py      # Adaptive pathing algorithm
│   │   │   ├── trace_generator.py # LLM reasoning traces
│   │   │   └── data_loader.py     # Taxonomy + catalog loader
│   │   └── main.py                # FastAPI app entrypoint
│   ├── requirements.txt
│   └── Dockerfile
├── frontend/
│   ├── src/
│   │   ├── components/
│   │   │   ├── UploadScreen.jsx
│   │   │   ├── ProcessingScreen.jsx
│   │   │   ├── ResultsDashboard.jsx
│   │   │   └── RoadmapView.jsx
│   │   ├── store/
│   │   │   └── useAppStore.js     # Zustand + localStorage persist
│   │   └── api/
│   │       └── client.js          # Axios API client
│   └── Dockerfile
├── docker-compose.yml
└── README.md

---

🔌 API Reference

Endpoint	Method	Description
POST /api/v1/analyze	POST	Upload resume + JD, returns job_id
GET /api/v1/status/{job_id}	GET	Poll job progress (0-100%)
GET /api/v1/results/{job_id}	GET	Get full pathway + gap report
GET /api/v1/trace/{job_id}/{skill_id}	GET	Get reasoning trace for a skill
GET /api/v1/stats	GET	System stats
GET /health	GET	Health check
GET /docs	GET	Interactive Swagger UI

---

📊 Evaluation Metrics

Metric	Target	Approach
Skill Extraction F1	> 0.82	Hybrid LLM + alias + NER
Gap Detection Recall	> 85%	3-layer matching pipeline
Pathway Validity	100%	Topological sort guarantee
Hallucination Rate	< 1%	Closed course catalog
E2E Latency (p95)	< 30s	Async background tasks

---

👥 Team

AI Apex - ARTPARK CodeForge Hackathon — 2024

---

📄 License

MIT License — built for ARTPARK CodeForge Hackathon