AI pipeline · Native iOS

Overeasy

From a cooking video to a recipe worth keeping.

I built a multi-stage AI pipeline that turns shared cooking videos into structured recipes, with a native iOS app for an offline recipe library and crowdsourced discovery.

My work

iOS app, extraction pipeline, model evaluation

Built with

SwiftUI, Python, FastAPI, Whisper

From video to library

1. TranscribeRecover speech from the shared video with Whisper.
2. ValidateCheck the evidence before extracting recipe details.
3. StructureProduce a consistent recipe the app can save offline.

Choosing the extraction model

I compared five LLMs on an 80-recipe text benchmark.

These figures describe structured text extraction, not end-to-end video processing or recipe accuracy.

Interactive learning · AI curriculum

Knute

Learn algorithms by working with them.

I built a native SwiftUI learning app with ten interactive problem formats and a multi-pass curriculum pipeline with human review before publication.

My work

Learning app, API architecture, curriculum pipeline

Built with

SwiftUI, TypeScript, Fastify, PostgreSQL, Zod

A flexible app architecture

Async/await networking and a protocol-based API layer let live and mock backends work interchangeably. Learning interactions can be developed independently of the live service.

A curriculum with several rounds of review

1. GenerateCreate candidate learning material.
2. ReviewSeven specialist reviewers assess correctness, pedagogy and prerequisites. Zod validates the structure.
3. ReconcileAdjudicate reviewer feedback before human review and publication.