SUYOG MAUNI

Diabetes Prediction System

Status: Completed & Deployed

A sophisticated machine learning-based web application that predicts the likelihood of diabetes in patients based on various health parameters. The system uses advanced ML algorithms to provide accurate predictions and insights.

Technologies: Python, Machine Learning, Flask, Scikit-learn, Pandas, NumPy

View Live Demo → View on GitHub →

AI Based Online Exam Proctoring System

Status: In Development

An intelligent proctoring system leveraging AI and computer vision to monitor online examinations. The system detects suspicious activities and ensures exam integrity through real-time video analysis and behavioral pattern recognition.

Technologies: Python, OpenCV, TensorFlow, Deep Learning, Flask, Computer Vision

Face Emotion Recognition System

Status: Completed & Deployed

A real-time facial emotion recognition system using a custom CNN (v5) with residual blocks trained on 48×48 grayscale face images. Detects 7 emotions — Happy, Sad, Angry, Surprise, Fear, Disgust, Neutral — from live webcam feed with temporal smoothing to reduce prediction flicker. Features a Flask web UI with live browser camera streaming, emotion distribution charts, and per-face confidence bars.

Model: Custom CNN v5 with residual blocks, 7-class emotion classification

Technologies: Python, TensorFlow, OpenCV, Flask, Flask-CORS, Haar Cascade

View on GitHub →

NetGuard — Network Threat Analyzer

Status: Completed & Deployed

A real-time network threat detection system powered by an ensemble ML model (Random Forest + Histogram Gradient Boosting) trained on the CICIDS2017 dataset with 2.8M+ flows. Features a live WebSocket dashboard with traffic monitoring, threat classification, protocol analysis, and automated IP blocking.

Model Accuracy: 99.47% across 7 threat classes (DDoS, Port Scan, Brute Force, Botnet, Web Attack, SQL Injection)

Technologies: Python, Scikit-learn, Flask, Socket.IO, Scapy, Chart.js, Docker

View Live Demo → View on GitHub →

AI Developer Hiring Suite

Status: Completed & Deployed

A full-stack NLP-powered recruitment tool that ranks engineering candidates by semantic fit using sentence-transformers (cosine similarity on 384-dim embeddings) and performs keyword gap analysis against job descriptions. Features an AI-powered code review system via Gemini 1.5 Flash that grades candidate code A–F with bug detection, security analysis, and refactoring suggestions.

NLP Model: all-MiniLM-L6-v2 — semantic similarity ranking across multiple PDF resumes

Technologies: Python, Flask, sentence-transformers, Google Gemini API, PyMuPDF, Scikit-learn

View Live Demo → View on GitHub →

AI Quiz Tutor

Status: Completed & Deployed

A full-stack AI-powered quiz platform that transforms any PDF into an interactive quiz. Upload lecture notes or textbooks and get automatically generated questions with difficulty levels, instant answer feedback, performance tracking, and per-difficulty accuracy stats — all powered by Groq LLaMA 3.1.

AI Model: LLaMA 3.1 8B via Groq API — generates contextual MCQs from extracted PDF chunks

Technologies: React, FastAPI, PostgreSQL, Groq API, JWT Auth, Python, Vite

View on GitHub →

Nepali Handwritten Character Recognition

Status: Completed & Deployed

A deep learning system for recognising handwritten Nepali (Devanagari) characters across 78 classes — 36 consonants, 10 numerals, and 12 vowels. Built with a custom ResNet + CBAM attention architecture trained on dual GPU using MirroredStrategy, achieving 97.37% TTA test accuracy. Features an interactive web canvas where users can draw any character and get real-time predictions with top-5 confidence scores.

Model: Custom ResNet + CBAM dual-pooling head, 11.4M parameters, 8× TTA ensemble inference

Technologies: Python, TensorFlow 2.13, Keras, OpenCV, Flask, Docker, Render

View on GitHub →