Projects

Full-stack AI workflow that reconciles meeting transcripts into evolving project action items with human-in-the-loop approval.

PythonFastAPISQLAlchemyPydanticPostgreSQL/SQLiteReactOpenAIDockerDocker ComposeREST API

Architected a proposal-based AI reconciliation engine that compares new meeting transcripts against persistent project state.
Designed a human-in-the-loop workflow separating AI 'proposal' from 'apply' to prevent direct LLM state mutation.
Built typed REST endpoints in FastAPI for project CRUD, action retrieval, and AI reconciliation workflows.
Implemented structured JSON validation of LLM responses using Pydantic to enforce deterministic data contracts.
Persisted evolving action state using SQLAlchemy ORM with relational modeling.
Developed a React frontend with side-by-side proposal comparison and approval interface.
Containerized the entire system using Docker Compose for single-command full-stack deployment.
Designed the system with clear separation of concerns: HTTP layer, business logic, AI layer, and persistence layer.

Serverless architecture for synchronized multi-sensor health data streaming.

React NativeAWS LambdaAPI GatewayWebSocketsAWS CognitoSupabase

Built a real-time patient monitoring system that synchronizes multiple physiological data streams into a single mobile interface.
Implemented a serverless backend using AWS Lambda and API Gateway with WebSocket streaming for low-latency data updates.
Developed secure authentication using AWS Cognito to protect patient data and manage device connections.
Designed a React Native mobile app to visualize synchronized sensor streams with responsive charts and real-time updates.
Impact: Demonstrates a scalable architecture for healthcare monitoring systems capable of integrating multiple medical sensors and supporting real-time clinical decision-making.

Track workouts, meals, and progress with a clean API and typed data model.

Next.jsTypeScriptPrismaPostgreSQLTailwind

Pairwise comparison system for building ranked lists efficiently.

ReactViteJavaScript

Built an interactive ranking system where users compare songs head-to-head to generate a personalized ordered list.
Implemented a binary-search style insertion algorithm that minimizes comparisons while placing songs in the correct rank.
Developed a dynamic React interface with ranked and unranked queues and a comparison view for fast decision making.
Structured a modular dataset for albums and artwork, enabling easy expansion to other artists or media categories.

This site — dark editor aesthetic, responsive layout, and content components.

Next.jsTypeScriptTailwind CSS