Field notes
I try to write up what I build, partly to think more clearly, partly so the next person does not have to start from scratch.
RedInk: A Multi-Agent Grading Assistant
RedInk is a multi-agent grading system I built using LangGraph. It uses specialized AI agents to parse rubrics, set standards, evaluate submissions, and generate feedback, with a mandatory human review step before anything is finalized.
Deja Vu: Near-Duplicate Video Detection with Deep Learning
For my Deep Learning course, I built a near-duplicate video detection system using CLIP and ResNet-50. This post documents the process of figuring out how to catch re-uploaded banned videos, even after edits like cropping, overlays, and reaction layouts.
Agentic Evolution: From Patent Data to Career Intelligence
For our DTSC 5082 group project, we built a system that uses USPTO patent data, financial indicators, and job-market signals to generate AI career intelligence reports. This post walks through what we learned along the way.
Cognitive Canvas : Exploring the Model Context Protocol (MCP).
Cognitive Canvas is an MCP (Model Context Protocol) Server that acts as a persistent memory layer for Claude Desktop and Copilot. It allows the AI to save URLs, categorize content, and generate quizzes for active recall.
Auto-Bench : Benchmarking Multi-Tool LLM Agents
Auto Bench is a reproducible benchmarking framework that evaluates multi-tool LLM agents by measuring their process (tool selection, reasoning, action sequencing), not just outcomes.
Final Eye Blink Tracker Implementation
This post is the final part of a series documenting my journey into building an eye blink tracker using PyTorch. It covers the complete implementation of the iTracker model, including data handling, model architecture, training pipeline, and performance evaluation.
PyTorch Eye Blink Tracker Pt.2
This post is the second part of a series documenting my journey into building an eye blink tracker using PyTorch. It covers the implementation of a Convolutional Neural Network (CNN) model, data handling techniques, and the use of the MNIST dataset for training and evaluation.
PyTorch Eye Blink Tracker Pt.1
This post is the first part of a series documenting my journey into building an eye blink tracker using PyTorch. It covers the fundamentals of PyTorch, including tensor operations, automatic differentiation, and the basics of neural networks, setting the stage for implementing a deep learning model for eye blink detection.
Global CO₂ Emissions Predictions: A Machine Learning Approach
This post details my journey into predicting global CO₂ emissions using machine learning techniques. I explored a Kaggle dataset spanning over 200 years and covering 190+ countries, applying time-series forecasting methods to understand historical patterns and predict future emissions.
Predictive modeling
This post details my journey into predictive modeling, focusing on housing market trends. I explored a global housing market dataset from Kaggle, applied machine learning techniques to forecast sales trends, and learned how to communicate insights through data visualization.
Affordable Education in Europe
This post presents a data visualization project analyzing education expenses across European countries, focusing on affordability and quality. It explores tuition fees, living costs, and education quality rankings to identify the best value for students seeking higher education in Europe.
Web scraping with Beautiful Soup
This post details my experience learning web scraping using Beautiful Soup and Requests in Python. I focused on extracting structured data from the World Economic Forum's Wikipedia page, navigating HTML tables, and handling inconsistencies in data extraction.
Building Database Management System Using SQL
A comprehensive guide to building a Hotel Management Database System using SQL, covering schema design, data population, stored procedures, query optimization, and performance tuning.