Kashmith.

My Projects

Final Year Project
Self-Adaptive Non-Intrusive Load Monitoring Using Deep Learning
  • Developed a novel NILM model to adapt to changes in power consumption patterns, particularly due to aging appliances.
  • Integrated deep learning techniques with transfer learning and pseudolabeling for autonomous adjustment to evolving power usage.
  • Utilized synthetic data generation and advanced neural network architectures for training and validation.
  • Achieved over 97% accuracy in disaggregating power consumption for a three-phase refrigerator over a six-year period.
  • Contributed to filling a significant gap in NILM literature and advancing energy management systems.
Industry Project
Tobu - Social Media Application
  • Developed the backend for a social media app integrated with large language models (LLMs) to enhance user interaction and content recommendation.
  • Implemented features for user images and conversation analysis using machine learning techniques.
  • Worked with LLMs to automate content creation and leveraged PySpark to build data pipelines for user data analysis and feature optimization.
Industry Project
Energy Consumption Prediction System
  • Designed and deployed an end-to-end energy forecasting pipeline using MLflow for experiment tracking, Apache Airflow for automated data workflows.
  • Built production-ready API services using FastAPI and Docker for real-time energy prediction and dashboard visualization.
  • Built an interactive Streamlit dashboard to visualize historical consumption, forecasted trends, and key model performance metrics.
Industry Project
CashTrack - Smart Expense Tracking Application
  • Developed a smart expense tracking application with beautiful analytics and insights using a modern React frontend with TypeScript, Tailwind CSS, and shadcn-ui components.
  • Built a powerful Jaseci backend with RESTful API architecture for data management and business logic implementation.
  • Implemented comprehensive features including expense/income tracking, data visualization with Victory charts, user profile management, and category-based organization.
Industry Project
Deep Research Assistant - Agentic AI Application
  • Built a production-grade agentic application for answering analytical questions using LangGraph for multi-step reasoning and workflow orchestration.
  • Integrated web search and document retrieval capabilities with Chroma vector database for efficient vector search and information retrieval.
  • Developed user-friendly Streamlit interface with comprehensive Docker containerization for easy deployment and horizontal scaling.
Industry Project
FriendZone - AI-Powered Memory Sharing Platform
  • Developed a full-stack AI-powered platform that allows users to capture, organize, and share memories through intelligent conversational interfaces.
  • Implemented automatic image analysis using OpenAI API to extract contextual information (who, what, where, when) from uploaded photos.
  • Built backend using jaclang and jac-cloud framework with comprehensive authentication, memory management, and social networking features.