Inclusive hair product recommendation web app that helps users build a hair profile, avoid unwanted ingredients, and receive explainable matches based on fit, budget, and hair goals.
This project combines a multi-page frontend with recommendation logic, product scoring, user profile persistence, and a grounded assistant experience. It stands out because the product is not just a quiz. It turns hair type, porosity, density, and shopping constraints into clear recommendations people can actually use.
- JavaScript
- HTML
- CSS
- Bootstrap
- Matches products by hair type, porosity, density, budget, and styling goals instead of vague one-size-fits-all advice.
- Explains why products fit and flags ingredient concerns rather than returning black-box recommendations.
- Includes a grounded assistant flow, automated tests, and deployment-ready Vercel setup.
Interactive housing-affordability experience that tests whether a recent graduate can realistically buy a starter home.
The repo combines dataset provenance, transform scripts, schema thinking, and interaction design to turn a broad economic question into an explorable product.
- JavaScript
- HTML
- CSS
- Data Visualization
- Uses BLS, Zillow, and Freddie Mac data to frame the affordability question.
- Lets users change assumptions with dynamic controls instead of reading a static chart.
- Documents limitations and next-step model improvements clearly.
Next.js counselor-style chat system that uses server-side AI tool calling and MCP architecture to guide users through orchestrator design.
This project combines a modern web app stack with orchestration patterns: App Router delivery, server-only model access, MCP tool boundaries, grounded counseling sources, and CI-backed quality gates. It stands out because the product is not just chat. It is a structured decision-support interface for designing human-centered AI orchestrators.
- TypeScript
- Next.js
- React
- Tailwind CSS
- Routes chat through a server-side API so model keys stay off the client.
- Uses MCP client-server patterns to ground responses with counselor sources and tool boundaries.
- Includes production-minded quality gates with linting, formatting, tests, Husky hooks, and CI.
AI-powered emotional music discovery platform that helps users find songs through mood, lyrical meaning, themes, and semantic similarity instead of searching only by title or artist.
This project combines Flask delivery, NLP-driven semantic search, embeddings, ChromaDB retrieval, Spotify integration, and a polished discovery interface to make music exploration feel emotional, searchable, and explainable. It stands out because it treats music discovery like a meaning-first search problem rather than a basic catalog lookup.
- Python
- Flask
- Sentence Transformers
- ChromaDB
- Lets users search by feelings, moods, themes, and personal experiences instead of exact song titles.
- Uses embeddings and vector search to return semantically related songs with lyric-based emotional interpretation.
- Connects discovery to Spotify metadata, artwork, and listening links through a polished, mood-driven interface.
FastAPI-based dataset cleaning application that combines a deterministic Pandas pipeline with retrieval-grounded explanations and a smart cleaning assistant for messy CSV workflows.
This project turns dataset cleaning into a guided product experience instead of a one-off script. It combines FastAPI delivery, deterministic mode-aware cleaning rules, RAG-backed explanations, and exportable reports so users can understand what changed, why it changed, and which cleaning tradeoffs fit their workflow.
- Python
- FastAPI
- Pandas
- ChromaDB
- Uploads messy CSVs, profiles data issues, and applies deterministic cleaning rules instead of black-box transformations.
- Supports workflow-specific modes for visualization, modeling, reporting, and exploration with different cleaning tradeoffs.
- Adds grounded assistant guidance and downloadable reports so users can review what changed and why.
Next.js blog with an AI consultant flow that turns voice input into publishable markdown content.
This build moves beyond a basic blog by combining auth, client-side transcription, LLM structuring, human approval, and GitHub-backed publishing into one workflow.
- TypeScript
- Next.js
- Firebase
- GitHub Actions
- Transforms recorded voice into structured post content.
- Keeps a human approval step before publishing.
- Uses repository automation so content updates can flow directly into deployment.
Streamlit app that turns live AI funding news into a structured, readable newsletter.
This is a focused automation pipeline: fetch source material, extract funding signals, shape the output, and display it through a lightweight web interface.
- Python
- Streamlit
- Feedparser
- Regex
- Pulls AI-related news from a live RSS source.
- Extracts funding amounts, round types, and investor names.
- Packages the result as a repeatable newsletter workflow rather than a one-off script.
Low-cost portfolio proof of concept built for GitHub Pages with content managed from a single source of truth.
This project matters because it shows product constraints clearly: zero-cost hosting, maintainability, redistribution, and AI-friendly content updates.
- Uses one content file to keep updates simple and non-technical.
- Optimized for free hosting and easy reuse.
- Structured to support future AI-assisted editing workflows.
Early local AI environment setup work captured in a minimal repository.
This repo is intentionally small, but it still signals hands-on experimentation with local model tooling and development containers.
- Captures local environment setup work rather than polished application code.
- Shows willingness to test local AI workflows early.
- Useful as supporting evidence of experimentation, not as the hero project.
