01 / ANALYSIS
Product analysis
- Maturity
- Live
- Confidence
- Verified
- Verified
- 2026-07-11
- Commercial relation
- Not sponsored
Problem solved
Restaurant menus often leave diners without a visual reference for unfamiliar dishes. MenuGen uses a menu photo to generate comparable dish references, addressing the concrete step of understanding what to order.
Why it was selected
MenuGen turns a restaurant menu photo into reference images for unfamiliar dishes. Karpathy says Cursor and Claude 3.7 generated the code, then documents the real friction around deployment, environment variables, Clerk, Stripe, rate limits, and stale APIs.
What is genuinely novel
Before payments, model slow work as saved, recoverable, retryable jobs instead of making one long request carry the product.
Best for
Independent builders taking an AI prototype into authentication, payments, and third-party APIs
Editorial evaluation
KEEP DISCOVERING
A few more products worth noticing
A small set selected by product shape and profile similarity, with independent verification records.
HealthSync
Harsh Patel
What it solves
Health data, sleep, water, steps, meal records, and voice notes are often split across devices and apps, making trends and next steps difficult to see together. HealthSync tries to bring those inputs into one dashboard.
Dropling
Uneed publisher donaldinos / Duckino s.r.o.
What it solves
Learning apps often bury review behind a destination users must remember to open. Dropling tries to bring questions to the Lock Screen users already see, then closes the loop with notification pacing, a Live Activity, lesson lists, and answer statistics.
Pennen
Uneed publisher IshaanRawat / independent maker Ishaan Rawat
What it solves
General note and journal apps can turn private reflection into infinite documents, streak maintenance, or AI analysis. Pennen tries to retain the boundedness and physical act of paper journaling while adding private iCloud sync.
02 / WHAT STANDS OUT
Three product decisions worth noticing
AI imagery
Before payments, model slow work as saved, recoverable, retryable jobs instead of making one long request carry the product.
Production postmortem
Before payments, model slow work as saved, recoverable, retryable jobs instead of making one long request carry the product.
Product insight worth carrying forward
Before payments, model slow work as saved, recoverable, retryable jobs instead of making one long request carry the product.
03 / DO NOT COPY
What not to copy
04 / LIMITS & RISKS
Limits and risks
The first version had no database or job queue, lost results on refresh, and could time out on long menus. Rate limits, stale docs, authentication, and payment mapping caused failures. Generated images cannot establish portion, ingredients, or allergy safety.
Compared with alternatives
Unlike a local-only demo it reached a public domain, auth, and payments; unlike a mature consumer app it still lacks robust persistence, async work, and recovery. The value is how clearly it exposes that gap.
05 / EVIDENCE
Visual evidence
06 / SOURCE
Source and verification
- Original author
- Andrej Karpathy
- Country / region
- United States
- Maturity
- Live
- Confidence
- Verified
- Discovered
- 2025-04-27
- Last verified
- 2026-07-11
What this review checked
We verified the first-party build log, public product entry, and the original result screenshot. It maps menu items to generated references while explicitly warning that the images are not the restaurant's actual dishes.
Ready to explore further?
Open the product, or return to the original source to verify the details.
ONE-TAP FEEDBACK
Did this profile help you judge the product?
One tap. No writing required.