Hello, I'm Gyandeep Mishra.

The Predictive Analytics in Autism Spectrum Disorder (ASD) project focuses on using artificial intelligence and deep learning techniques to support the early and accurate diagnosis of autism.

Motivation & Problem

Autism Spectrum Disorder is a neurodevelopmental condition that affects communication, behavior, and social interaction. Early diagnosis plays a critical role in enabling timely intervention. However, traditional diagnosis methods largely depend on behavioral observations and expert judgment, which can be time-consuming, subjective, and prone to inconsistencies, often leading to delayed diagnosis or false positives.

The Solution

The motivation behind this project was to design a data-driven and objective diagnostic support system that can assist healthcare professionals by improving diagnostic accuracy and reducing uncertainty. By leveraging deep learning models, the system aims to identify autism-related patterns at an early stage, making the diagnostic process faster and more reliable.

Technical Implementation

In this project, a Convolutional Neural Network (CNN) was developed using TensorFlow and Keras. The model was trained on 5,000+ datasets, which include relevant features associated with Autism Spectrum Disorder. CNNs are particularly effective in identifying complex and non-linear patterns in data, making them well-suited for medical prediction tasks.

Results & Impact

After training and evaluation, the model achieved a prediction accuracy of 94%, demonstrating strong diagnostic capability. Additionally, the system successfully reduced false positive rates by 15%, which is critical in avoiding unnecessary stress, misdiagnosis, and incorrect treatment plans.

By improving diagnostic precision, this project contributes to early autism detection, enabling healthcare professionals and caregivers to initiate timely intervention strategies. The system is intended to act as a decision-support tool, assisting clinicians rather than replacing professional medical judgment.

Overall, this project highlights the potential of AI-driven predictive analytics in healthcare, specifically in neurodevelopmental disorder diagnosis, enhancing accuracy, consistency, and efficiency in medical decision-making.

The Face Recognition Attendance System using ESP32 is an automated attendance management solution designed to replace traditional manual and biometric attendance methods, which are often time-consuming, error-prone, and inconvenient.

Motivation

In educational institutions and organizations, taking attendance manually consumes valuable time and can lead to proxy attendance or incorrect record keeping. The motivation behind this project was to develop a contactless, fast, and accurate attendance system that leverages computer vision and embedded systems. Especially after the COVID-19 period, the need for touch-free and automated systems became more important for hygiene, efficiency, and scalability.

How It Works

The system uses an ESP32 camera module to capture facial images of individuals in real time. These images are processed using Python and OpenCV, where facial features are detected and compared against a pre-stored database using the face_recognition library. Once a face is successfully recognized, the system automatically marks the attendance of the individual.

Performance & Tech Stack

The system is capable of recognizing 100+ faces in under 2 seconds, making it suitable for real-time classroom or office environments. Attendance data is stored and managed efficiently using Pandas and NumPy, enabling structured record-keeping and easy data analysis.

Impact

During evaluation, the system achieved an overall recognition accuracy of 95%, ensuring reliable performance in real-world conditions. By automating the attendance process, the system saves approximately 15 minutes per session, significantly improving productivity for instructors and administrators.

This project demonstrates how embedded hardware (ESP32) combined with computer vision and machine learning techniques can solve everyday administrative problems, highlighting the practical application of AI in building smart, efficient, and scalable systems.

Overview: A high-performance Deep Packet Inspection (DPI) engine designed to analyze, classify, and filter network traffic in real time using a scalable multi-threaded architecture.

The Problem

Traditional network monitoring systems rely on basic packet inspection techniques that only analyze packet headers, making them ineffective for identifying modern applications, encrypted traffic patterns, and malicious activities. Additionally, single-threaded systems struggle to handle high-throughput network traffic, leading to performance bottlenecks and delayed analysis.

The Solution

To address these limitations, this project implements a multi-threaded DPI engine capable of inspecting packet payloads in depth while maintaining high performance. The system processes PCAP-based network traffic and distributes workloads efficiently across multiple threads to ensure scalability and real-time analysis.

Core Features

• Deep Packet Inspection (DPI): Analyzes both packet headers and payloads to identify applications and traffic patterns.
• Multi-threaded Architecture: Uses thread-safe queues and load balancing to process large volumes of network packets in parallel.
• Connection Tracking: Maintains stateful information of network flows for accurate traffic classification.
• Rule-based Filtering System: Allows blocking or filtering of traffic based on predefined rules (e.g., domain or protocol-based).
• SNI Extraction (TLS Analysis): Extracts Server Name Indication (SNI) from encrypted traffic to identify domains without decryption.
• PCAP Processing Engine: Reads and processes offline packet capture files for traffic analysis and testing.

System Architecture

The system follows a pipeline-based design:

PCAP Input → Packet Parser → Load Balancer → Worker Threads → DPI Engine → Rule Engine → Output

This architecture ensures efficient packet distribution and high throughput while maintaining accuracy in inspection.

Technical Contribution

• Designed a scalable multi-threaded pipeline for packet processing
• Implemented lock-free / thread-safe queue mechanisms for inter-thread communication
• Built modular components such as: Packet parser, Load balancer, Rule manager, Connection tracker
• Optimized packet flow handling to minimize latency and maximize throughput

Impact

• Enables real-time traffic classification and filtering
• Improves performance and scalability compared to single-threaded DPI systems
• Demonstrates practical implementation of network security concepts used in ISPs and enterprise firewalls
• Provides a foundation for advanced applications like intrusion detection systems (IDS) and network monitoring tools

Tech Stack

• Language: C++
• Concepts: Multi-threading, Networking, Systems Design
• Libraries/Tools: PCAP processing, CMake
• Core Components: Thread-safe queues, Load balancer, DPI engine, Rule manager