Bias Fairness Analyzer
A dataset auditing platform for analyzing bias, imbalance, and classification performance across NLP and structured datasets.
Built a dataset auditing platform supporting both NLP and tabular classification datasets
Implemented bias and imbalance detection including class distribution skew, demographic imbalance, and sensitive attribute analysis
Developed NLP analysis pipelines to detect linguistic bias, sentiment imbalance, toxicity signals, and gendered language patterns
Integrated classification model evaluation to report performance metrics (accuracy, precision, recall, F1-score) alongside dataset diagnostics
Designed a bias risk scoring and reporting system generating structured insights on fairness risks and dataset quality
Created an end-to-end automated pipeline
Multimodal Local LLM AI Assistant
A privacy-first AI assistant running entirely on-device using a local Large Language Model (LLM).
Built a multimodal AI assistant supporting voice input, OCR-based screen understanding, and contextual chat
Designed a modular AI pipeline integrating speech recognition, text extraction, and LLM reasoning
Implemented real-time streaming responses using a FastAPI backend for low-latency interaction
Developed a browser extension integration for contextual, in-page AI assistance
Ensured 100% local inference, eliminating external API dependency and enhancing data privacy
Tech Stack: Python, FastAPI, React, Local LLM, OCR, Speech Recognition, Browser Extension
Bias Analysis Platform
Developed a machine learning analysis platform to detect demographic and linguistic bias in structured and textual datasets. Automated dataset evaluation pipelines using Python, FastAPI, and Scikit-learn, generating accuracy, precision, recall, and F1-score reports. Delivered a React-based dashboard to visualize bias metrics and generate dataset risk assessment summaries.
Multimodal LLM AI Agent
Engineered a multimodal AI assistant supporting voice input, OCR-based screen understanding, and contextual chat interactions. Integrated Tesseract OCR, SpeechRecognition, and locally hosted LLMs through Ollama for privacy-focused AI processing. Streamed real-time responses through FastAPI and Server-Sent Events.
API Architecture & Testing Platform
Constructed a full-stack platform to visualize API request flows and backend service dependencies. Implemented secure JWT-based authentication and RESTful API architecture using Node.js and Express. Enabled developers to trace requests, inspect responses, and analyze backend behavior across distributed API calls.
Health Disease Prediction using Machine Learning Techniques: An Analysis
Conducted research on machine learning models for healthcare disease prediction using structured datasets. Applied classification algorithms and data preprocessing techniques, and evaluated model performance using standard ML metrics and comparative analysis. Published or presented at ICAS 2025.
