Hello and welcome to my demo apps portal! I am a data scientist, ex-quant, and developer who enjoys building intelligent, data-driven applications. My work often sits at the intersection of analytics, machine learning, and modern AI — from real-time scoring systems to RAG-powered agents and custom index builders.
This page showcases a selection of hands-on demos and experimental prototypes I’ve built while exploring new technologies, frameworks, and ideas. Each project highlights a different aspect of what I enjoy doing: designing systems, solving complex data problems, and transforming technical concepts into interactive products you can try directly in your browser.
Projects
Index Maker RAG
RAG-powered agent that builds custom financial indices from natural-language instructions. It retrieves relevant market data, interprets your constraints, and produces an index on demand — complete with visualizations that show how the index evolves over time. You can experiment with sector, factor, or rules-based index ideas and see them computed and plotted instantly. A simple preview of how LLMs can automate index design and turn complex quantitative workflows into interactive, intuitive tools.
CIFAR-10 - Image Recognition
Before transformers and embeddings reshaped the field, convolutional neural networks were the foundation of image recognition. This “old-school” project shows how a classical CNN architecture is applied on the CIFAR-10 dataset — a collection of 60,000 images spanning 10 categories such as airplanes, frogs, and horses. The model uses a transfer-learning approach built on MobileNetV2, originally trained on ImageNet, with the final 50 layers unfrozen and fine-tuned for CIFAR-10. The result is a classifier that reaches an accuracy of 89%.
Wealth Mate - RAG for Beginner Investors, powered by Youtube captions
WealthMate is a RAG system built on embedded captions from a popular finance YouTube channel (Plan Bagel). It’s designed to give beginner investors credible answers by grounding every response in real video content — not guesses or hallucinations. The app retrieves answers to user questions along with the YouTube video from which the answer is generated, as well as audio responses. While similar results might be achievable with a well-crafted ChatGPT prompt or even a Google search, the key difference is that this RAG system keeps every response anchored to a clearly defined and verifiable knowledge base. This ensures consistency, transparency, and reliability — essential when teaching the fundamentals of personal finance.
Robo Reviews - Product Insights
Robo Reviews is an automated pipeline that transforms 65,000+ Amazon product reviews into clean, production-ready summaries for e-commerce sites. It extracts sentiment using a local twitter-roberta-base-sentiment-latest model, infers product categories with gpt-4o-mini, and generates final article-style summaries using the local meta-llama/Llama-3.2-1B-Instruct model. The system identifies key product groups, ranks the best and worst items within each category, and condenses customer opinions into clear, neutral summaries. With a single run, it outputs structured JSON files ready to plug into a product-review website.
