SRM Career Catalyst

Intelligent Resume Analysis Engine

SRM Career Catalyst is an advanced, data-driven backend system designed to analyze student resumes against job descriptions. Unlike standard keyword matchers, this system leverages a Knowledge Genome built from the resumes of successfully placed alumni to provide context-aware, actionable feedback.

The platform integrates Traditional NLP, Unsupervised Machine Learning, and Generative AI to uncover real-world placement patterns and guide students toward improved placement outcomes.

---

🧠 Key Idea

"Compare a student not to a job description alone, but to the profiles of alumni who actually got placed."

---

🏗️ System Architecture

graph TD
    A[Raw Alumni Resume PDFs] --> B["PDF & Text Extraction (PyPDF + Regex)"]
    
    B --> C["NLP Entity Recognition (spaCy - Skills, Orgs, Dates)"]
    B --> D["Achievement Analysis (GenAI - Metrics Extraction)"]
    B --> E["Text Vectorization (SentenceTransformers)"]
    
    E --> F["Career Archetype Discovery (K-Means Clustering)"]
    
    C --> G[("Metadata Store")]
    D --> G
    F --> H[("FAISS Vector Store + Metadata")]
    
    I[Student Resume PDF] --> J[Resume Extraction Node]
    K[Job Description Text] --> L[RAG Retrieval Node]
    
    L -->|Semantic Query| H
    H -->|Relevant Archetypes & Alumni Examples| M[GenAI Reasoning Node]
    J -->|Structured Resume Data| M
    
    M --> N["Actionable Feedback Report (Markdown Output)"]
    
    style A fill:#e1f5ff
    style I fill:#e1f5ff
    style K fill:#e1f5ff
    style N fill:#d4edda
    style G fill:#fff3cd
    style H fill:#fff3cd

Loading

---

🔍 Phase 1: Offline Data Pipeline

This phase builds the intelligence layer of the system.

1. Data Extraction

• Converts unstructured alumni resume PDFs into structured text.
• Tools: PyPDF, Regex

2. Entity Recognition

• Extracts skills, organizations, roles, and timelines.
• Model: spaCy (en_core_web_lg)

3. Achievement Analysis

• Uses GenAI to identify quantifiable achievements.
• Example:
  • "Reduced latency by 20ms" → Speed / Performance
• Model: gpt-4o-mini (via GitHub AI + Azure SDK)

4. Archetype Discovery

• Converts resumes into dense embeddings.
• Uses K-Means clustering to discover natural career archetypes.
• No manual labels required initially.

5. Vector Store Creation

• Embeddings and metadata indexed using FAISS.
• Enables high-speed semantic retrieval during inference.

---

⚡ Phase 2: Online Inference Engine

This phase exposes the system as a REST API.

Workflow

1. Request Handling

   • Resume PDF + Job Description received via FastAPI.

2. Orchestration

   • Workflow managed using LangGraph:
     • Extraction → Retrieval → Reasoning

3. RAG (Retrieval-Augmented Generation)

   • Retrieves the most relevant alumni archetypes from FAISS.

4. Synthesis

   • GenAI compares student profile against:
     • Successful alumni patterns
     • Required job skills
     • Quantified achievements

5. Output

   • Markdown-formatted feedback
   • Skill gaps
   • Missing metrics
   • Resume improvement suggestions

---

🛠️ Technology Stack

Core

• Language: Python 3.9+
• API: FastAPI, Uvicorn
• Workflow: LangGraph, LangChain

AI & ML

• Generative AI: GitHub AI (openai/gpt-4o-mini)
• NLP: spaCy (NER)
• ML: Scikit-Learn (K-Means, TF-IDF)
• Embeddings: Sentence-Transformers (all-MiniLM-L6-v2)

Storage & Processing

• Vector DB: FAISS (CPU)
• Data: Pandas, NumPy
• PDF Parsing: PyPDF, Regex

---

📂 Project Structure

Placement-Project/
├── processed_data/
│   ├── structured_resumes.csv
│   ├── resumes_with_metrics.csv
│   ├── clustered_resumes.csv
│   ├── embeddings.npy
│   └── archetype_insights.json
├── vector_store/
│   ├── srm_resumes.index
│   └── srm_resumes.pkl
├── data/
│   └── raw_resumes/
├── extract_resume_data.py
├── analyze_achievements.py
├── cluster_resumes.py
├── label_archetypes.py
├── build_vector_store.py
├── services.py
├── graph.py
├── models.py
├── main.py
├── requirements.txt
└── .env

---

🚀 Setup & Installation

1. Prerequisites

• Python 3.9+
• GitHub Account (for Personal Access Token)

2. Environment Setup

python -m venv .venv

# Windows
.\.venv\Scripts\Activate.ps1

# Mac/Linux
source .venv/bin/activate

3. Install Dependencies

pip install -r requirements.txt

4. Download NLP Models

python -m spacy download en_core_web_lg

5. Configuration

Create a .env file in the project root:

GITHUB_TOKEN="your_github_pat_token_here"

---

⚙️ Execution Guide (Building the Knowledge Base)

Step 1: Extract Resume Data

python extract_resume_data.py

Step 2: Analyze Achievements (GenAI)

python analyze_achievements.py

Step 3: Clustering

Run 1: Generate elbow plot

python cluster_resumes.py

• Inspect elbow_plot.png
• Update OPTIMAL_K in the script

Run 2: Final clustering

python cluster_resumes.py

Step 4: Label Archetypes

Run 1: Inspect keywords

python label_archetypes.py

• Update ARCHETYPE_LABELS

Run 2: Save labeled insights

python label_archetypes.py

Step 5: Build Vector Store

python build_vector_store.py

---

⚡ Running the API Server

python -m uvicorn main:app --reload

Server runs at:

http://127.0.0.1:8000

---

📡 API Documentation

Endpoint

POST /analyze

Parameters

• job_description (Query Parameter): Full job description text
• resume_file (Body): Student resume PDF

Response

• Streaming Markdown text
• Alumni archetype comparison
• Skill gap analysis
• Actionable resume recommendations

---

🎯 Outcome

SRM Career Catalyst moves beyond resume screening and into career intelligence, enabling students to align their profiles with real-world placement success patterns.

---

📜 License

For academic and research use. Extendable for institutional deployment.