Efficient Object Segmentation for Recycling Applications 🔗

• Reduced real-time video object segmentation processing time by 57.1% using a novel skip-frame approach with object tracking and Q-TinySAM zero-shot segmentation.

Microbe-Drug Association Prediction 🔗

• Achieved a 95.2% precision-recall score through a self-supervised learning-based model using multi-modal inputs.
• Created metapath-aggregated heterogeneous graphs to identify complex, salient associations between gut microbes and drugs.

Automated Literature Survey with Graph Retrieval-Augmented Generation (GraphRAG) 🔗

• Architected a reliable, LLM-powered chatbot that improved reliability by 81% using a knowledge-graph-based RAG framework.
• Enhanced citation coverage and visualized retrieved scientific paper networks with LlamaIndex and Neo4j.
• Attained a quantitative F1 BERTScore of 71% and a qualitative RAGAS score of 85% for generating literature reviews.

KAIZEN Community Health-Booth 🔗

Smart City Initiative | Runners-up, GITEX 2023 - Largest Technology Expo in Dubai

• Deployed a real-time vital monitoring system, achieving seamless sensor-software integrations with Raspberry Pi.
• Engineered a modular data processing pipeline enabling data ingestion via wired and wireless connections.
• Enabled early detection and proactive healthcare measures via predictive analytics for obesity risk and epidemic forecasting based on vitals monitoring for a community.

Smart Software Merge Conflict Resolution using Deep Learning 🔗

• Conducted a comprehensive study analyzing and improving existing models’ prediction capabilities on tabular data.
• Attained an 81% prediction confidence in streamlining the decision-making process and managing merge conflicts, thereby boosting developer efficiency using Gradient Boosted Decision Trees with Neural Networks.

Recipe Curator 🔗

• Conceptualized a custom algorithm to curate recipes from pantry-specific ingredients.
• Designed a full-stack web application with an intuitive UI using HTML, CSS, JS, and back-end integration with Python (Flask) and MySQL.
• Web-scraped 50+ popular food blogs using Selenium WebDriver to automate recipe collection. Processed and standardized data with an NLP toolkit, enhancing best match reliability by up to 85%.

Document Parser - Electronic Thesis & Dissertation Search Engine 🔗

• Enhanced content extraction using YOLO object detection to generate digital documents.
• Coordinated cross-team indexing and integration of results into the knowledge graph and UI, while optimizing data handling and batch processing with Apache Kafka.