Projects
RAG
Users can ask AI questions about whether food ingredients are allowed on therapeutic diets.
The app uses Gemini 2.5 Flash-lite to read context acquired from a vector database. The vector database was built from a JSON database of food ingredients I built for Meadow Mentor. My goal here was to learn the workflow for building RAG features into AI apps.
Ideally, start with clean, structured data. Use an embedding model to turn the data into embeddings. Store the embeddings in a vector database. Use a similarity search to find the most relevant documents. Use a model to generate a response.
Tech stack includes python Fast API, Gemini 2.5 Flash-lite, Chromadb, Next.js/React, Docker, Google Cloud, Google Gemini.
On-the-Fly AI Analysis of System Prompts
As AI converses with users, it analyses how effective the system prompt is guiding its responses against a rubric. This allows for rapid iteration on system prompts to ensure user safety, tone, and domain accuracy.
Tech stack includes Python Fast API, LangChain, Gemini 2.5 series, and Next.js.
Meadow Mentor
An AI-powered health application designed to provide personalized wellness guidance and support. This is an ongoing project where I do the full product management work, talking to users, analyzing data, and engineering solutions.
This full-stack app includes React 19, Material UI, Tailwind, Node.JS, Express, MongoDB, Sanity, Docker, Google Cloud, Google Gemini.
RecipeWreck
An absurdist take on AI generated recipes. This shows off how AI can generate structred JSON recipe content and have that rendered on the page without AI text conversation. Also includes a landing page designed and written by Gemini 2.5 Pro. I dare you not to laugh!
Node.js Express AI Documentation Guidelines
A standard for AI-generated JSDoc for Node.js Express applications, ensuring clarity and consistency.