kiran karlapati

Job nest

- Developed JobNest, a full-stack job portal connecting recruiters and job seekers through a centralized platform. - Built the backend using Java 17 and Spring Boot to provide REST APIs for core job portal functionalities. - Implemented user registration and login with Spring Security and JWT authentication. - Designed role-based access control for three roles: Admin, Recruiter, and Candidate. - Developed job posting, browsing, searching, and management functionalities for recruiters and candidates. - Implemented job application workflows, allowing candidates to apply for jobs and track application statuses. - Enabled recruiters to manage their job postings and view applications submitted for their jobs. - Integrated MySQL, Spring Data JPA, and Hibernate for database storage and entity relationship management. - Followed a layered architecture using Controller, Service, and Repository layers to separate API handling, business logic, and data access. - Integrated Swagger/OpenAPI for API documentation and testing, with a React-based frontend for user interaction.

DeepFake detection using multi model

- Developed DeepTrace, an explainable AI-based deepfake detection system for analyzing suspicious videos. - Integrated EfficientNetB4 to classify video frames as real or potentially fake. - Implemented Grad-CAM heatmaps to visualize regions influencing the model’s predictions. - Built a custom audio–lip synchronization algorithm using Librosa and MediaPipe FaceMesh. - Applied Gaussian smoothing and Pearson correlation to identify potential audio–visual mismatches. - Designed a smart frame-sampling strategy and selective heatmap generation to optimize CPU performance. - Integrated a locally running Llama 3.2 (3B) model using Ollama to generate forensic explanations. - Developed a Streamlit frontend and FastAPI backend for video uploads and analysis. - Generated downloadable PDF forensic reports containing detection results, confidence scores, and visual evidence. - Designed the system to operate locally using open-source tools, reducing cloud dependency and improving accessibility.

Fresh Sense AI

- Developed FreshSense AI, an AI-powered web application that detects food freshness and classifies food images as fresh, stale, or spoiled. - Integrated a pretrained image classification model through the Hugging Face Inference API for AI-based food image analysis. - Built an intuitive web interface allowing users to upload food images and view prediction results. - Implemented image processing and API integration to send uploaded images for model inference. - Designed a classification workflow to identify potential food spoilage based on visual characteristics. - Integrated MongoDB to store relevant application data and analysis records. - Developed backend functionality to handle image submissions, communicate with the AI inference service, and process prediction responses. - Presented classification results through a user-friendly interface to make food freshness assessment accessible. - Used a modular architecture to separate frontend interactions, backend processing, and AI inference. - Designed the system to support practical use cases such as household food checks, grocery quality assessment, and food-waste reduction.