Back to products
FocusChef thumbnail

FocusChef

B2CBeta

Cooking without the overwhelm.

One step at a time, with AI that actually helps when things go wrong. Type what you want to cook and get a complete, personalized recipe in seconds.

GeminiCookingLLM ChainReact/TS

The Problem

Traditional recipes create cognitive overload—walls of text, vague instructions, no help with parallelization, and zero support when things go wrong. For anyone with executive dysfunction or decision fatigue, cooking becomes anxiety-inducing.

The Approach

Recipes are not static documents. FocusChef treats them as dynamic, living guides that adapt to you. Paste a URL, drop in text, or just describe what you want—then scale servings, swap ingredients, and adjust to your skill level. Single-step focus with persistent timers and real-time AI help.

Features

Concept-to-recipe generation—type 'spicy miso ramen' and get a complete recipe
Any recipe transformed—paste a URL or text, AI extracts and enhances it
One-step-at-a-time interface with no scrolling or position tracking
Persistent floating timers that survive navigation and refresh
Real-time AI help for troubleshooting mid-cook
Invisible parallelization—steps interleaved so wait time is productive

Technical Highlights

7-stage recipe pipeline with Zod validation at each boundary

Two-phase chain: Phase 1 blocks for intro, Phase 2 runs in background

Timestamp-based timers survive browser refresh and device switches

PWA session resume with schema migration

Learnings

1

Decompose complex AI tasks into focused stages with validation boundaries

2

'Prompt engineering' is really 'pipeline engineering'

3

Parallelization must be invisible or it fails—users shouldn't think about it

4

Micro-rewards work better than end-of-task celebrations