Skillence

AI-Powered Career Intelligence Platform

Skillence is a full-stack web application that combines resume parsing, AI career recommendations, job market analytics, campus placement management, and interview coaching into a single unified platform. Built as a final-year B.Tech project.

Features

Resume Intelligence

AI-Powered Resume Parsing — Upload PDF/DOCX resumes. Azure AI Document Intelligence extracts text, then Google Gemini structures it into categorized data (skills, experience, education, projects, certifications).
Fallback Parser — spaCy + NLTK regex-based parser activates automatically if Azure/Gemini are unavailable.
SHA-256 Deduplication — Prevents re-processing of identical uploads.
Profile Editor — Review and manually edit parsed resume data before saving.

Career Path Recommendation

3-Stage AI Pipeline — (1) Map user skills to a 692-skill vocabulary via exact match, aliases, and fuzzy regex. (2) Dot-product pre-filter narrows 894 O*NET occupations to top 30 candidates in ~1ms. (3) Google Gemini ranks and scores the top 10 with explanations.
3-Phase Learning Roadmaps — Personalized skill gap analysis using O*NET Skills, Technology Skills, and Knowledge databases. Synonym-aware matching with 70% coverage threshold. ML-enhanced skill recommendations.
Progress Tracking — Save career paths, track learning milestones, mark skills as completed with checkpoint persistence.

Job Market Analytics

Rich Dashboard — Analyze ~30,000 job postings with advanced filtering (location, industry, experience level, salary range, company size, employment type).
Trendiness Scoring — Composite score: recency (40%) + salary (30%) + benefits (20%) + growth (10%).
AI Insights — Gemini-powered market analysis and career advice.
ML Salary Prediction — Neural network predicts salaries based on skills, experience, location, industry, and more.
Data Export — Download filtered results as CSV or JSON.

Skill Libraries

Rich Learning Content — Detailed skill profiles including prerequisites, use cases, learning difficulty, and top courses from platforms like Udemy and Coursera.
YouTube API Integration — Fetching and displaying top educational videos for each skill using the YouTube Data API v3.
Interactive Roadmap — Visual, trackable learning roadmap with expandable phases, topics, and completion checkboxes.
Dynamic Resources — Fetching relevant external resources (MDN, GeeksforGeeks, W3Schools, freeCodeCamp) for specific roadmap topics using DuckDuckGo search and static fallbacks.
Bookmark System — Robust system for users to save skills with status tracking (Interested, Learning, Completed) and course progress percentage.
Practice Platforms — Direct links to relevant coding, AI, and cloud practice sites.

Job Offer Evaluator

Side-by-Side Comparison — Compare two job offers with salary charts (Chart.js), cost-of-living analysis via Gemini, and interactive Leaflet maps with Nominatim geocoding.
Market Context — Adzuna API integration for salary benchmarks across 25+ countries.

Campus Placement System

Placement Cell Dashboard (admin role) — Create and manage company drives, upload/parse job descriptions (Gemini + keyword fallback), manage course catalogs, shortlist students algorithmically.
Student Portal — Browse drives (upcoming/expired/not-eligible tabs), upload grade history PDFs, view eligibility and match scores, apply and track applications.
Matching Engine — Pure algorithmic scoring (zero LLM): total = required_skills×0.70 + preferred_skills×0.15 + resume_bonus×0.10 + cgpa_bonus×0.05. Eligibility gate checks 10th%, 12th%, CGPA, and active backlogs.
Grade History Parsing — VIT-style PDF parser extracts courses, grades (S=10 to F=0), credits, and CGPA. Handles retakes by keeping best grade.
Course→Skill Mapping — Curriculum PDFs parsed and mapped to canonical skills via Gemini AI.

AI Chatbot

Career Guidance — Multi-turn conversational AI powered by Gemini. Context-aware with user profile data. Includes platform navigation guidance.
Speech Support — Web Speech API for speech-to-text input and text-to-speech output.
Session Persistence — Chat history stored in localStorage across page refreshes.

Reflection Engine (Interview Coach)

Interview Diagnostic — AI-powered interview coaching using a structured "MistakeLoop" methodology.
Contextual Coaching — Incorporates user profile and career goals for personalized feedback.

ML Models

Skill Recommender — Denoising Autoencoder trained on O*NET occupation–skill data. Recommends skills based on reconstruction confidence, suppressing 40+ generic/non-actionable terms.
Salary Predictor — MLP with skip connections trained on ~30K job postings. Encodes skills (binary vector), experience, location, industry, company size, education level, and employment type.
NumPy-Only Inference — Both models are trained with PyTorch offline, exported to .npz, and run in production via pure NumPy to avoid PyTorch DLL crashes on Windows.

Tech Stack

Layer	Technology
Backend	FastAPI 0.104.1, Python 3.8+, Uvicorn
Frontend	React 19.1, Vite 7, pnpm
Database	MongoDB Atlas (async Motor 3.3 driver)
AI Services	Google Gemini 2.5 Flash, Azure AI Document Intelligence
ML Training	PyTorch 2.x (offline only)
ML Inference	Pure NumPy (no PyTorch at runtime)
Auth	JWT (HS256) + bcrypt + Google OAuth 2.0
Career Data	O*NET (894 occupations, 692-skill vocabulary)
Charts	Recharts (job trends), Chart.js (offer evaluator)
Maps	Leaflet + react-leaflet + Nominatim geocoding
Icons	Lucide React
Animations	Motion library
WebGL	OGL (neural network background)
Speech	Web Speech API (STT/TTS)
HTTP	Fetch API (general), Axios (chatbot)

Architecture

┌─────────────────────────────────────────────────────────────┐
│                     Frontend (React 19)                      │
│  Port 3000 · Vite 7 · pnpm · react-router-dom v7           │
│                                                              │
│  ┌──────────┐ ┌────────────┐ ┌──────────┐ ┌──────────────┐ │
│  │  Resume   │ │  Career    │ │  Job     │ │  Placement   │ │
│  │ Dashboard │ │  Recommend │ │  Trends  │ │  (Student/   │ │
│  │          │ │            │ │          │ │   Admin)     │ │
│  └──────────┘ └────────────┘ └──────────┘ └──────────────┘ │
│  ┌──────────┐ ┌────────────┐ ┌──────────┐ ┌──────────────┐ │
│  │  Chatbot │ │  Reflection│ │  Job     │ │   Profile    │ │
│  │          │ │  Engine    │ │  Offer   │ │   Page       │ │
│  └──────────┘ └────────────┘ └──────────┘ └──────────────┘ │
└──────────────────────┬──────────────────────────────────────┘
                       │ REST API (fetch / axios)
┌──────────────────────▼──────────────────────────────────────┐
│                  Backend (FastAPI)                            │
│  Port 8000 · 10 Routers · Async Motor · JWT Auth            │
│                                                              │
│  Routers:  auth · resume · profile · career_path · chatbot  │
│            job_trends · ml_predictions · placement_cell      │
│            student_placement · skills                        │
│                                                              │
│  Services: 14 service modules (auth, resume parsing,        │
│            career matching, learning plans, job trends,      │
│            placement matching, JD parsing, academics, etc.)  │
│                                                              │
│  ML:       NumPy-only inference (skill recommender +         │
│            salary predictor) loaded lazily on first request  │
└──────────────────────┬──────────────────────────────────────┘
                       │
          ┌────────────┼────────────────┐
          ▼            ▼                ▼
   ┌────────────┐ ┌──────────┐ ┌────────────────┐
   │  MongoDB   │ │  Gemini  │ │  Azure AI Doc  │
   │  Atlas     │ │  2.5     │ │  Intelligence  │
   │            │ │  Flash   │ │                │
   └────────────┘ └──────────┘ └────────────────┘

Getting Started

Prerequisites

Python 3.8+
Node.js 16+ with pnpm
MongoDB Atlas instance (or local MongoDB)
API Keys: Google Gemini, Azure AI Document Intelligence
Optional API Keys: O*NET Web Services, JSearch (RapidAPI), Adzuna, Google OAuth, YouTube Data API v3

Environment Variables

Create a .env file in the project root (not inside backend/ or frontend/):

# Database
MONGODB_URI=mongodb+srv://...

# JWT Authentication
SECRET_KEY=your-secret-key
ALGORITHM=HS256
ACCESS_TOKEN_EXPIRE_MINUTES=30

# Azure AI Document Intelligence
AZURE_ENDPOINT=https://your-resource.cognitiveservices.azure.com/
AZURE_API_KEY=your-azure-key

# Google Gemini AI
GEMINI_API=your-gemini-api-key
GEMINI_ENDPOINT=https://generativelanguage.googleapis.com/v1beta/models/gemini-2.5-flash

# Email (Password Reset)
EMAIL_HOST=smtp.gmail.com
EMAIL_PORT=587
EMAIL_HOST_USER=your-email@gmail.com
EMAIL_HOST_PASSWORD=your-app-password
FRONTEND_URL=http://localhost:3000

# CORS (Backend)
# Comma-separated values
CORS_ORIGINS=http://localhost:3000,http://127.0.0.1:3000
# Regex for preview deployments (Vercel)
CORS_ORIGIN_REGEX=https://.*\\.vercel\\.app

# Data Enrichment (Optional - for ML training only)
ONET_API_KEY=your-onet-key
JSEARCH_API_KEY=your-jsearch-key

# Frontend (VITE_ prefix required)
VITE_API_URL=http://localhost:8000
VITE_GOOGLE_CLIENT_ID=your-google-oauth-client-id
VITE_ADZUNA_API_KEY=your-adzuna-key
VITE_ADZUNA_APP_ID=your-adzuna-app-id
VITE_GEMINI_API_KEY=your-gemini-key-for-frontend

Installation & Running

Backend:

cd backend
pip install -r requirements.txt
uvicorn main:app --reload --port 8000

Frontend:

cd frontend
pnpm install
pnpm run dev

The frontend runs at http://localhost:3000 and the backend API at http://localhost:8000.

Deployment (Vercel + Render)

This repository is set up for split deployment:

Frontend: frontend/ on Vercel
Backend: backend/ on Render

1) Deploy backend to Render (GitHub import)

Import this repo into Render.
Use the existing render.yaml blueprint, or create a Web Service manually with:
- Root Directory: backend
- Build Command: pip install -r requirements.txt
- Start Command: uvicorn main:app --host 0.0.0.0 --port $PORT
Set environment variables in Render (prompted by sync: false in blueprint), especially:
- MONGODB_URI, SECRET_KEY, GEMINI_API, AZURE_ENDPOINT, AZURE_API_KEY
- FRONTEND_URL = your Vercel production URL (e.g. https://your-app.vercel.app)
- CORS_ORIGINS = comma-separated Vercel + local origins
After deploy, copy your Render backend URL (e.g. https://your-backend.onrender.com).

2) Deploy frontend to Vercel (GitHub import)

Import this repo into Vercel.
Set Root Directory to frontend.
Framework preset: Vite (auto-detected).
Ensure environment variable:
- VITE_API_URL = your Render backend URL (no trailing slash)
SPA routing fallback is configured in frontend/vercel.json.

3) Local + production compatibility

Local dev defaults still work (VITE_API_URL defaults to http://localhost:8000).
Backend CORS now supports both local origins and configured deployment origins.
Password reset email links now use FRONTEND_URL instead of hardcoded localhost.

Project Structure

/
├── .env                             # All environment variables (root level)
├── .github/copilot-instructions.md  # AI agent context (comprehensive)
├── package.json                     # Root package.json
│
├── backend/
│   ├── main.py                      # FastAPI entrypoint, 10 routers
│   ├── requirements.txt             # Python dependencies
│   ├── run_salary_pipeline.py       # CLI: salary model training pipeline
│   ├── seed_course_catalog.py       # CLI: curriculum PDF → MongoDB seed
│   └── app/
│       ├── database.py              # Async Motor client
│       ├── models/                  # Pydantic models (user, resume, placement, roadmap)
│       ├── routers/                 # 10 API routers
│       ├── services/                # 14 business logic services
│       ├── utils/                   # JWT, role auth, email
│       ├── data/                    # skill_taxonomy.json
│       ├── career_data/             # O*NET JSON + Excel + job CSVs
│       └── ml/                      # ML pipeline (models, training, inference, data)
│
├── frontend/
│   ├── package.json                 # React 19, Vite 7
│   ├── vite.config.js               # Port 3000, envDir: root
│   └── src/
│       ├── App.jsx                  # Routes, ProtectedRoute, RoleRoute
│       ├── main.jsx                 # GoogleOAuthProvider, theme init
│       └── components/              # 23+ components organized by feature
│
└── docs/                            # Documentation files

API Reference

Authentication — `/api/auth`

Method	Endpoint	Auth	Description
POST	`/register`	—	Create account (student or placement_cell)
POST	`/login`	—	Email/password login → JWT
POST	`/google-login`	—	Google OAuth login (Gmail only)
GET	`/verify`	Bearer	Verify JWT validity
GET	`/me`	Bearer	Get current user info
POST	`/forgot-password`	—	Send reset email
POST	`/reset-password`	—	Reset with token

Resume — `/api/resume`

Method	Endpoint	Auth	Description
POST	`/upload`	Bearer	Upload resume file
POST	`/parse`	Bearer	Parse uploaded resume (Azure AI + Gemini)
GET	`/{user_id}`	Bearer	Get parsed resume data

Profile — `/api/profile`

Method	Endpoint	Auth	Description
POST	`/save`	Bearer	Save profile data
GET	`/`	Bearer	Get user profile
PUT	`/update`	Bearer	Update profile

Career Path — `/api/career-path`

Method	Endpoint	Auth	Description
POST	`/recommendations`	Bearer	Get AI career recommendations (top 10)
POST	`/save-career`	Bearer	Save selected career path
GET	`/learning-plan`	Bearer	Generate 3-phase learning roadmap
PUT	`/learning-plan`	Bearer	Update learning progress
POST	`/update-learned-skill`	Bearer	Mark skill as learned

Chatbot — `/api/chatbot`

Method	Endpoint	Auth	Description
POST	`/chat`	Optional	Career guidance conversation

Reflection Engine — `/api/reflection-engine`

Method	Endpoint	Auth	Description
POST	`/chat`	Optional	Interview coaching conversation

Skills — `/api/skills`

Method	Endpoint	Auth	Description
GET	`/`	—	List minimal skill info for browsing
GET	`/search`	—	Search skills by name or category
GET	`/{skill_id}`	—	Get full details of a specific skill
GET	`/youtube/{skill_id}`	—	Fetch top YouTube educational videos
GET	`/{skill_id}/resources`	—	Fetch external resources for a roadmap topic
GET	`/user/activity`	Bearer	Get user's saved skills and progress
POST	`/user/activity`	Bearer	Update saved skills, progress, and status

Job Trends — `/api/job-trends`

Method	Endpoint	Auth	Description
GET	`/jobs`	—	List jobs with filters
GET	`/analysis/{job_title}`	—	Job analysis
GET	`/skills/{job_title}`	—	Skill demand for job
GET	`/overview`	—	Market overview
GET	`/trends/{job_title}`	—	Time-series trends
GET	`/experience-distribution/{job_title}`	—	Experience breakdown
GET	`/filter-options`	—	Available filter values
GET	`/ai-insights`	—	AI market insights
GET	`/ai-insights-gemini`	—	Gemini-specific insights
GET	`/detailed-analysis/{job_title}`	—	Comprehensive analysis
GET	`/cache-info`	—	Cache status
POST	`/clear-cache`	—	Clear data cache
GET	`/export/csv`	—	Export as CSV
GET	`/export/json`	—	Export as JSON
POST	`/predict-salary`	—	ML salary prediction

ML Predictions — `/api/ml`

Method	Endpoint	Auth	Description
POST	`/recommend-skills`	—	Skill recommendations from autoencoder
GET	`/health`	—	Model health check

Placement Cell — `/api/placement` (placement_cell role)

Method	Endpoint	Auth	Description
POST	`/drives`	Role	Create company drive
GET	`/drives`	Role	List all drives
GET	`/drives/{id}`	Role	Get drive details
PUT	`/drives/{id}`	Role	Update drive
DELETE	`/drives/{id}`	Role	Delete drive
POST	`/drives/{id}/jd`	Role	Upload/parse JD
GET	`/courses`	Role	List course catalog
POST	`/courses`	Role	Add course
PUT	`/courses/{id}`	Role	Edit course
DELETE	`/courses/{id}`	Role	Delete course
POST	`/curriculum/upload`	Role	Upload curriculum PDF
POST	`/drives/{id}/shortlist`	Role	Shortlist students
PUT	`/applications/{id}/status`	Role	Update application status

Student Placement — `/api/student/placement` (student role)

Method	Endpoint	Auth	Description
GET	`/academics`	Role	Get academic profile
POST	`/academics/grade-history`	Role	Upload grade history PDF
PUT	`/academics`	Role	Update 10th/12th%
GET	`/academics/skills`	Role	Get skill profile
GET	`/drives`	Role	Browse drives (with tabs)
GET	`/drives/{id}`	Role	Drive detail + match score
POST	`/apply/{drive_id}`	Role	Apply to drive
GET	`/applications`	Role	View applications
DELETE	`/applications/{id}`	Role	Withdraw application

ML Pipeline

Both models follow the same workflow: Train (PyTorch) → Export (.npz) → Infer (NumPy).

Skill Recommender

Architecture: Denoising Autoencoder — Encoder: V→256→128→64→32 (latent). Decoder mirrors. BatchNorm + LeakyReLU(0.2) + Dropout(0.3). Sigmoid output.
Data: Binary skill–occupation vectors from O*NET Excel files, occupation JSON, job CSVs, and cached API data.
Training: BCE loss, Adam optimizer, 30% input corruption, early stopping (patience=20).

Salary Predictor

Architecture: MLP with skip connections — Input→512→256→128→64→1. BatchNorm + ReLU + Dropout(0.3). He initialization.
Data: ~30K job postings. Features: binary skill vector, scaled experience, label-encoded categoricals, one-hot experience level.
Training: MSE loss, Adam, gradient clipping, early stopping (patience=25).

Running the ML Pipeline

cd backend

# Skill recommender
python -m app.ml.data.fetch_api_data          # Fetch O*NET + JSearch data (optional)
python -m app.ml.data.skill_data_processor     # Build vocabulary
python -m app.ml.training.train_skill_recommender  # Train
python -m app.ml.export_to_numpy               # Export to .npz

# Salary predictor (all-in-one)
python run_salary_pipeline.py
# Or with flags: --skip-processing --skip-training --skip-export --model lite|full

Course Catalog Seeding

cd backend
python seed_course_catalog.py --pdf ../curr.pdf

User Roles

Role	Access	Key Pages
student (default)	Resume upload, career recommendations, job trends, placement portal, chatbot	`/dashboard/resume`, `/career-path-recommendation`, `/campus-placement`
placement_cell	All student features + drive management, JD parsing, shortlisting, curriculum management	`/placement-dashboard`
Unauthenticated	Landing page, job trends, job offer evaluator, reflection engine, static pages	`/`, `/job-trends`, `/job-offer-evaluator`, `/reflection-engine`

External Services

Service	Purpose	Used In
Azure AI Document Intelligence	Resume text extraction (PDF/DOCX)	Backend: `azure_resume_parser.py`
Google Gemini 2.5 Flash	Career ranking, learning plans, chatbot, interview coaching, JD parsing, resume structuring, course mapping, cost-of-living	Backend (6+ services) + Frontend (Job Offer Evaluator)
*ONET Web Services**	Occupation data enrichment for ML training	Backend: `fetch_api_data.py`
JSearch (RapidAPI)	Job posting data for ML training	Backend: `fetch_api_data.py`
Adzuna API	Salary benchmarks, job market data	Frontend: `JobOfferEvaluator`
Nominatim (OSM)	Geocoding for map location picker	Frontend: `MapLocationPicker`
Google OAuth 2.0	Social login	Frontend → Backend
YouTube Data API v3	Fetching educational videos for skills	Backend: `youtube_service.py`
DuckDuckGo Search	Dynamic scraping of learning resources	Backend: `resource_fetcher.py`

Key Design Decisions

NumPy-Only Inference — PyTorch triggers c10.dll / WinError 1114 crashes when loaded inside async FastAPI on Windows. Models are trained offline with PyTorch, exported to .npz weight files, and the forward pass uses 4 pure-NumPy functions (_linear, _batchnorm, _leaky_relu/_relu, _sigmoid).
Single Root .env — All environment variables live in one .env at the project root. Vite uses envDir: '..' to access it. Backend services load it via dotenv with relative paths.
No Global State Management — Frontend uses React useState + localStorage only. No Redux, Context API, or Zustand. Auth tokens, chat sessions, and theme preferences are all in localStorage.
Gemini JSON Repair — Gemini responses frequently include trailing commas, truncated output, or markdown fences. Multiple services implement _try_repair_json() helpers.
Lazy Model Loading — ML models (SkillRecommender, SalaryPredictorInference) are loaded on first API request, not at server startup, to keep startup fast.
Service-Oriented Backend — Routers delegate to services; services own business logic. Exception: career_path.py has inline scoring logic alongside service calls (technical debt).
Dual Resume Parser — Primary: Azure AI + Gemini (high accuracy). Fallback: spaCy + NLTK + regex (works offline). Automatic failover.
Pure Algorithmic Placement Matching — The matching engine uses zero LLM calls. Weighted formula with eligibility gates ensures deterministic, fast, and explainable results.

Database Collections

Collection	Purpose
`users`	User accounts (email, hashed password, role, name)
`profiles`	Parsed resume data (contact, skills, experience, education, projects)
`resumes`	Resume metadata with SHA-256 file hashes
`company_drives`	Placement drives with JD, criteria, package info
`applications`	Student applications to drives with status tracking
`student_academics`	CGPA, 10th/12th%, courses, skills, grade history
`course_catalog`	Course → skill mappings from curriculum PDFs
`learning_roadmaps`	Saved career paths and learning plan progress
`user_skills`	Saved skills, progress, and status tracking for users

License

This project was built as a final-year B.Tech capstone project.

Name		Name	Last commit message	Last commit date
Latest commit History 107 Commits
.github		.github
backend		backend
docs		docs
frontend		frontend
.env.example		.env.example
.gitignore		.gitignore
README.md		README.md
REMEDIATION_GUIDE.md		REMEDIATION_GUIDE.md
SECURITY_AUDIT_SUMMARY.md		SECURITY_AUDIT_SUMMARY.md
SECURITY_FINDINGS.md		SECURITY_FINDINGS.md
package-lock.json		package-lock.json
package.json		package.json
render.yaml		render.yaml
test_skills.py		test_skills.py

Folders and files

Latest commit

History

Repository files navigation

Skillence

Features

Resume Intelligence

Career Path Recommendation

Job Market Analytics

Skill Libraries

Job Offer Evaluator

Campus Placement System

AI Chatbot

Reflection Engine (Interview Coach)

ML Models

Tech Stack

Architecture

Getting Started

Prerequisites

Environment Variables

Installation & Running

Deployment (Vercel + Render)

1) Deploy backend to Render (GitHub import)

2) Deploy frontend to Vercel (GitHub import)

3) Local + production compatibility

Project Structure

API Reference

Authentication — /api/auth

Resume — /api/resume

Profile — /api/profile

Career Path — /api/career-path

Chatbot — /api/chatbot

Reflection Engine — /api/reflection-engine

Skills — /api/skills

Job Trends — /api/job-trends

ML Predictions — /api/ml

Placement Cell — /api/placement (placement_cell role)

Student Placement — /api/student/placement (student role)

ML Pipeline

Skill Recommender

Salary Predictor

Running the ML Pipeline

Course Catalog Seeding

User Roles

External Services

Key Design Decisions

Database Collections

License

About

Resources

Uh oh!

Stars

Watchers

Forks

Releases

Packages 0

Uh oh!

Contributors

Uh oh!

Languages

Authentication — `/api/auth`

Resume — `/api/resume`

Profile — `/api/profile`

Career Path — `/api/career-path`

Chatbot — `/api/chatbot`

Reflection Engine — `/api/reflection-engine`

Skills — `/api/skills`

Job Trends — `/api/job-trends`

ML Predictions — `/api/ml`

Placement Cell — `/api/placement` (placement_cell role)

Student Placement — `/api/student/placement` (student role)

Packages