Explore our Projects
Discover cutting-edge technology projects that shape the future.
A Platform for Retrieval Augmented Generation (RAG) in Sports Medicine
Objective:
To develop a flexible Retrieval Augmented Generation (RAG) platform for the Sports Medicine domain that allows users to select any Large Language Model (LLM) and embedding method, experiment with different RAG approaches, and evaluate their design choices using an integrated performance assessment framework.
Description:
This project delivers a customizable RAG-based chatbot designed to handle complex queries over domain-specific documents, such as medical research papers and case studies in sports medicine. Users can select from various LLMs (e.g., OpenAI, Gemini, LLaMA 3, Gamma 2) and embedding techniques to optimize retrieval performance. The system supports multiple RAG strategies, including Naive RAG and advanced GraphRAG variants, enabling experimentation with diverse retrieval and reasoning workflows.
The platform incorporates agentic capabilities, allowing it to perform multi-step reasoning, query multiple data sources, and dynamically use tools and APIs to generate highly contextualized responses. A user-friendly Streamlit frontend enables real-time query input and results visualization, while a FastAPI backend ensures scalable, high-performance processing. Integrated with ARES (Automatic Retrieval Evaluation System), the platform continuously assesses relevance, accuracy, and coherence of generated answers, providing actionable feedback for system improvement. This approach empowers researchers, practitioners, and developers to fine-tune RAG systems for maximum effectiveness in sports medicine knowledge retrieval and decision support.
AI for Smart Homes: From Sunlight to Socket
Objective:
To develop a machine learning–driven system that accurately forecasts solar energy production and household electricity consumption, enabling smarter energy management through predictive analytics, web dashboards, and virtual assistant integration.
Description:
This project applies deep learning to forecast both solar energy production and household electricity consumption for smarter energy management. Using datasets from Malappuram (solar) and Delhi (consumption), models like ARIMAX, Prophet, and multivariate LSTM were compared, with LSTM delivering the best accuracy (Solar RMSE: 2.55 kWh, Consumption R²: 0.88). Data was enriched with NASA climate metrics and engineered features like seasonality and time-of-day patterns.
A proposed web-based dashboard will allow real-time visualization of forecasts, while an AI-powered virtual assistant will answer natural-language energy queries. This system aims to support households, utility providers, and sustainability initiatives by predicting trends, detecting anomalies, and optimizing energy usage.
AI for Video Analysis – Advertisements Tracking in NBA Game Videos
Objective:
To develop an AI-powered system that can automatically detect, track, and analyze advertisements in NBA game videos, providing accurate insights on ad exposure, frequency, and effectiveness for advertisers, broadcasters, and sports organizations.
Description:
This project leverages computer vision and deep learning techniques to monitor advertisements displayed during NBA games, including those on digital billboards, courtside banners, and other in-game surfaces. Using models such as YOLO or Faster R-CNN for ad detection, along with OCR to identify brand names and ad content, the system will classify and track ads based on location, brand, and visibility duration. Video processing tools like OpenCV and FFmpeg will be used for frame extraction and pre-processing, while data analytics will provide detailed reports on ad occurrences, exposure time, and effectiveness. The solution also supports real-time ad monitoring, enabling stakeholders to assess sponsorship value, optimize marketing strategies, and maximize revenue potential.
Document Analysis using Large Language Models (LLMs)
Objective:
To develop an intelligent, multi-agent document analysis platform powered by Large Language Models (LLMs) that can process a variety of document types and video content, extract key insights, and enable semantic search and interactive querying.
Description:
This project leverages LLMs to build a versatile document analysis system capable of handling diverse inputs, including PDFs, Word documents, plain text files, and YouTube videos. Uploaded files are parsed into meaningful chunks for precise and efficient processing, while YouTube videos are transcribed from audio to text for analysis. The processed content is stored in a vector database, enabling semantic search and Retrieval-Augmented Generation (RAG)-style querying.
