Semantic Candidate Matching Engine

AI-powered talent intelligence platform for semantic candidate discovery, hybrid ranking, and explainable hiring recommendations.

Live Demo

Streamlit Application
https://semantic-candidate-matching-engine-uxggpfryj2kq3kta4bxawr.streamlit.app/

GitHub Repository
https://github.com/reddyeswaranush/semantic-candidate-matching-engine

---

Problem Statement

Traditional resume screening systems rely heavily on keyword matching, causing highly qualified candidates to be overlooked when their profiles use different terminology.

This project uses semantic search, vector retrieval, and hybrid ranking to identify the most relevant candidates based on meaning rather than exact keyword matches.

---

Features

• Semantic candidate search using BGE embeddings
• FAISS vector similarity retrieval
• Hybrid 11-signal ranking engine
• Query-aware role detection
• Explainable candidate reasoning
• Candidate upload via CSV or Excel
• Automatic embedding generation
• Automatic FAISS index rebuilding
• Recruiter-friendly Streamlit dashboard
• CSV export for shortlisted candidates

---

Project Metrics

• Supports 100,000+ candidate profiles
• 384-dimensional BGE embeddings
• FAISS cosine similarity retrieval
• 11-signal hybrid ranking engine
• Explainable AI-powered recommendations
• Cloud deployment using Streamlit

---

Screenshots

Dashboard Overview

The recruiter dashboard provides a summary of the candidate database, embedding model, retrieval engine, and ranking system.

---

Candidate Search

Recruiters can paste a job description and configure the number of results to retrieve.

---

Search Results

The platform retrieves the most relevant candidates using semantic search and hybrid ranking.

Features demonstrated:

• Semantic retrieval using BGE embeddings
• FAISS vector similarity search
• Hybrid ranking engine
• Top candidate recommendations

---

Candidate Analysis

Each candidate receives a detailed breakdown of ranking signals and explainable reasoning.

Metrics include:

• Final Score
• Semantic Score
• Skill Match Score
• Experience Score
• Career Score
• Seniority Score
• Company Score

---

Candidate Upload Workflow

Recruiters can upload candidate profiles using CSV or Excel files. The system automatically generates embeddings and rebuilds the FAISS index.

---

Project Structure

semantic-candidate-matching-engine/
│
├── app.py
├── requirements.txt
├── README.md
├── migrate.py
├── setup.py
│
├── assets/
│   ├── dashboard.png
│   ├── search_interface.png
│   ├── search_results.png
│   ├── candidate_details.png
│   ├── upload_candidates.png
│   └── architecture.png
│
├── src/
│   ├── candidate_schema.py
│   ├── candidate_upload.py
│   ├── storage_manager.py
│   ├── full_embeddings.py
│   ├── full_retrieval.py
│   ├── ranking.py
│   ├── reasoning.py
│   └── submission_generator.py
│
├── data/
└── storage/

---

System Architecture

Candidate Retrieval Pipeline

Recruiter Job Description
            │
            ▼
    BGE Embeddings
            │
            ▼
   FAISS Vector Search
            │
            ▼
 Top Semantic Candidates
            │
            ▼
 Hybrid 11-Signal Ranking
            │
            ▼
 Explainable Recommendations

---

Tech Stack

Machine Learning

• BAAI/bge-small-en-v1.5
• Sentence Transformers
• NumPy

Vector Search

• FAISS IndexFlatIP
• Cosine Similarity Search

Backend

• Python
• JSON Storage Layer

Frontend

• Streamlit

Data Processing

• Pandas
• OpenPyXL

---

Candidate Search Pipeline

Step 1: Candidate Upload

Recruiters upload candidate data through CSV or Excel files.

Required fields:

• candidate_id
• current_title
• years_of_experience
• current_company
• skills

Step 2: Embedding Generation

Each candidate profile is transformed into a semantic representation using BGE embeddings.

Output:

• 384-dimensional embedding vector

Step 3: Vector Indexing

Candidate embeddings are stored inside a FAISS vector index for efficient retrieval.

Step 4: Semantic Retrieval

The job description is embedded and matched against candidate embeddings using cosine similarity.

Step 5: Hybrid Ranking

Candidates are ranked using:

• Semantic Score
• Skill Match Score
• Experience Score
• Career Score
• Title Match Score
• Behavior Score
• Seniority Score
• Company Score
• Role Profile Matching
• Query-Aware Skill Alignment
• Final Weighted Ranking

Step 6: Explainable Recommendations

Each recommendation includes recruiter-friendly reasoning explaining why the candidate was selected.

---

Example Use Cases

Machine Learning Hiring

• Python
• Machine Learning
• NLP
• FAISS
• RAG
• PyTorch

Generative AI Hiring

• LLMs
• RAG
• LangChain
• LlamaIndex
• Prompt Engineering
• Vector Databases

Data Analytics Hiring

• SQL
• Power BI
• Tableau
• Excel

Backend Hiring

• Python
• Java
• Docker
• Kubernetes
• PostgreSQL

---

Deployment

The application is deployed on Streamlit Cloud and supports:

• Candidate upload
• Semantic retrieval
• FAISS indexing
• Explainable ranking
• CSV export

---

Future Improvements

• Multi-company workspaces
• PostgreSQL storage backend
• Candidate profile enrichment
• LLM-powered recruiter summaries
• Analytics dashboard
• API integrations
• Authentication and role-based access control

---

Author

Reddy Eswar Anush

B.Tech Computer Science and Engineering
National Institute of Technology Agartala

GitHub
https://github.com/reddyeswaranush

LinkedIn
https://www.linkedin.com/in/reddy-eswar/

Project Repository
https://github.com/reddyeswaranush/semantic-candidate-matching-engine

Live Demo
https://semantic-candidate-matching-engine-uxggpfryj2kq3kta4bxawr.streamlit.app/