Avijit
Pakhira

Django-powered event management platform with user registration, event creation & booking, job applications, chatbot integration, and admin approval workflows. Supports customizable templates for weddings, conferences & workshops, plus email notifications and media handling for images and resumes.

ML-powered health analytics tool that forecasts diabetes risk from patient metrics like glucose level, BMI, and age. Features an interactive Streamlit web app for real-time predictions, a trained model pipeline, and a full Jupyter notebook covering data exploration, feature engineering, and model evaluation on the PIMA dataset.

Gradio-based exploratory data analysis dashboard that lets users upload any CSV and instantly get interactive charts, correlation heatmaps, distribution plots, and statistical summaries. Includes user authentication and a clean, no-code interface ideal for quick data profiling.

Dual-model AgriTech system — an LSTM/CNN-LSTM stack predicts soil moisture trends over time, while Random Forest, Decision Tree & Logistic Regression ensemble recommends the optimal crop based on soil nutrients and climate data. Empowers farmers with data-driven irrigation and planting decisions to maximize yield.

Python-based voice assistant for Windows that converts spoken commands to text via speech recognition and responds through Windows SAPI text-to-speech. Navigates YouTube, Wikipedia, and Google hands-free, integrates OpenAI API for intelligent responses, and manages local media playback — a fully voice-driven desktop AI companion.

Real-time object detection pipeline with Non-Maximum Suppression & dynamic confidence thresholding. Processes live video feed, draws labeled bounding boxes, and is optimized for edge device deployment with reduced inference latency.

End-to-end computer vision pipeline that detects filled bubbles, compares them against an answer key, and auto-grades OMR sheets in real-time. Eliminates manual checking with sub-second evaluation speed and high accuracy across varying sheet qualities.

Deep learning model for precise autonomous lane boundary detection, built during the Pinnacle internship. Uses Hough transform + CNN for robust lane tracking under varied lighting and road conditions, designed for self-driving car simulation pipelines.

NodeMCU + ultrasonic sensor system connected to Blynk cloud for real-time tank level monitoring and automated pump control. Mobile push alerts, historical level logging, and smart refilling logic — reducing household water waste by 30%.