AI, ML and Data Science Projects

These are a few applied AI and data science projects where I take messy data, build models, evaluate them carefully, and package the results into product-style experiences. I focus on the full loop, defining what success looks like, choosing the right metrics, and designing the system so it is measurable when used. Funnily enough, I built this portfolio too :P. Turns out a lot of the “free” site builders are not actually free anymore :/

Spendlark Money

Spendlark Money is a personal finance insights app that turns raw transaction statements into a clean monthly story of where your money went. It ingests statements and organizes transactions into understandable categories so you can see spending patterns without manual tracking. It highlights month-over-month changes, flags unusual spikes, and surfaces subscription waste that quietly adds up. It also gives forward-looking guidance so you can plan smarter for the next month. The experience is powered by machine learning for categorization and pattern detection, and an LLM layer that turns insights into simple, guided coaching.

Financial Analytics Machine Learning Spend Classification Anomaly Detection Forecasting LLM application

PMS Mood Compass

PMS Mood Compass is a cycle and wellbeing tracker that goes beyond logging and helps you understand patterns over time. You track symptoms and lifestyle signals in a simple daily flow, and the app turns that history into clear insights you can actually act on. It helps you spot which habits tend to correlate with better or worse days instead of leaving you with a long list of entries. It also provides a short-term outlook based on your past tracking, so you know what to expect and can plan ahead. The intelligence is backed by machine learning that learns from your own logs and surfaces interpretable trends.

Health Analytics Machine Learning Time Series Forecasting FastAPI Predictive Modeling Product Insights

StudyBuddyAI

StudyBuddyAI is an AI education platform that teaches machine learning and modern AI in a way that builds real intuition, not just definitions. Lessons break concepts down step by step so the experience feels structured, beginner-friendly, and still technically solid. It also includes an interactive tutor so learners can ask questions while studying and get answers that stay aligned to the lesson they are on, making the experience feel closer to having a personal instructor than searching random explanations online. The tutor is LLM-backed, uses retrieval grounding to stay consistent with the course content, and incorporates a LangSmith-based evaluation workflow to trace interactions, inspect failure cases, compare prompt variants, and improve response reliability over time.

Learning Platform EdTech LLM application Retrieval Augmented Generation LangChain LangSmith Personalization

Fraud Detection Analysis

This is an end-to-end machine learning project built on transaction-level data to detect suspicious financial activity. It starts with exploratory analysis to understand fraud patterns across transaction type, amount, and timing, then builds interpretable features that capture how risky behavior shows up in payment flows. The project compares classification models, handles class imbalance carefully, and tunes the decision threshold to balance fraud capture against false alerts. It also includes model interpretation and prediction scoring so the output is not just accurate, but usable in a practical fraud-review setting.

Fraud Analytics Machine Learning Classification Modeling Feature Engineering Risk Modeling Model Interpretation Precision-Recall Analysis

Behavioral Anomaly Detection for Digital Platforms

This is an end-to-end machine learning project built on user activity data to identify suspicious or abnormal behavior patterns. It starts with exploratory analysis to compare normal and unusual engagement, then builds interpretable behavioral features around repetition, concentration, session activity, burstiness, and shared infrastructure signals. The project combines anomaly detection with supervised modeling, evaluates threshold tradeoffs between risk capture and false positives, and includes account-level interpretation so the output is useful for integrity monitoring, trust and safety workflows, and product decision-making.

Behavioral Analytics Anomaly Detection Machine Learning Feature Engineering Classification Modeling Risk Scoring Model Interpretation Threshold Optimization Product Analytics Trust and Safety Streamlit

Retail Customer Intelligence Platform

This project looks at how customers shop over time and uses transaction data to uncover patterns in retention, spending, and product relationships. It brings together segmentation, cohort analysis, value prediction, and recommendations to make the data more useful for understanding customer behavior and supporting practical retail decisions.

Retail Analytics Customer Behavior Segmentation Predictive Modeling Recommendations Streamlit

YouTube Comment Spam Detection & Review Queue Optimization

This project looks at how spam comments can be detected and routed through a moderation workflow using public YouTube comment data. It brings together text classification, threshold analysis, error analysis, and review queue simulation to show how moderation teams can balance spam detection quality with false alarms and reviewer capacity. The project is presented through an interactive Streamlit app that helps explore model tradeoffs and operational decisions in a way that feels closer to a real Trust & Safety analytics workflow.

Trust & Safety Moderation Systems Spam Detection NLP Precision-Recall Review Queue Simulation Streamlit

SQL Sprint

SQL Sprint is a realistic SQL interview practice platform designed to mirror real screening rounds. It uses business-style datasets across domains so the questions feel like what companies actually ask. Your queries are evaluated using real query outputs, which keeps scoring objective and repeatable. The platform also explains mistakes clearly so you learn the patterns behind the problems, not just the final answer. It includes an LLM-based helper to guide debugging and reasoning while keeping the evaluation grounded in SQL correctness.

SQL LangChain Data Evaluation Full-Stack Web App Evaluation Frameworks

GlowKind

GlowKind is an AI-assisted beauty product search and comparison app that makes clean shopping feel simple instead of overwhelming. You can search a product and immediately see ingredient and packaging signals explained in plain language. It highlights ingredient watch-outs, summarizes safety and sustainability signals, and makes it easier to compare products within the same category. It also recommends cleaner alternatives that are actually relevant to what you are shopping for. The scoring is backed by machine learning so product comparisons stay consistent rather than relying on vague marketing claims.

Consumer Product Sustainability Machine Learning Search Recommendations Ranking and Scoring