https://img.leopedia.io/DQmYu5pwurVEvqgbatqGh1xBEtMS7BTkBTurE5oAQceFbFC/MenuGen%20presentation.png

Andrej Karpathy’s "MenuGen" web app embodies a fresh, minimalist approach to apps development through AI-driven tools that Karpathy himself dubbed "vibe coding". Let's first explore MenuGen – that could be adopted by some Lions 😀 - before tackling Karpathy’s background, the vibe coding phenomenon, and why it has resonated so much with a lot of developers and non coding creators alike.

# «Bon APP-etit»: the Purposes Behind MenuGen

In Andrej Karpathy’s words: «I basically had this problem where I show up at a restaurant, I read through the menu and I have no idea what any of the things are… I need pictures». [see Andrej Karpathy’s keynote from June 17, 2025 at AI Startup School in San Francisco: «Software in the era of AI» - https://inv.nadeko.net/watch?v=LCEmiRjPEtQ]

Some hours of vibe coding later, the web app was running that same day on Andrej’s phone!:

https://www.menugen.app/

**All users get $3 in credits for free upon signing up.**

At its core, MenuGen does exactly what its name suggests: it allows us to take a picture of a menu (its dishes names), and it creates images that match each one of the listed dishes. Hence, its audience is two-fold: people like you and me, when we’re confronted with an exotic menu where only a small percentage of the dishes comes with a picture, as well as chefs or restaurant owners who’d wish to refresh their menu with attractive photos. Whether creating a hipster brunch spot in Brooklyn, a traditional Italian trattoria, or an experimental French molecular gastronomy setting, the app captures the essence of different restaurant styles with surprising accuracy. 

 https://img.leopedia.io/DQmUiU32F4MLUW4mo1pWYgEXYJi86pZPQqz7dXBxphXN3es/MenuGen%20homepage%20-%20Turn%20Menus%20into%20Magic.png

Another purpose of MenuGen, however, extends beyond menu creation. It serves as a demonstration of how relatively simple generative AI apps can produce content that feels genuine, coherent, and culturally aware. For developers and designers, MenuGen provides inspiration regarding how AI can be leveraged for creative projects through focused applicationsthat require maintaining a consistent tone in a narrow, well-defined domain. Instead of generic, sprawling outputs, MenuGen creates flavorful, cohesive menus that feel authentic to the restaurant style it emulates. 

For the broader tech community, it represents a playful yet powerful example of how AI can understand and replicate cultural patterns in a way that feels almost human.

# How MenuGen Works: Minimal Architecture, Maximum Vibe
MenuGen achieves its charm and effective output through a lightweight, highly optimized AI setup that runs entirely in the browser. The core mechanic involves calling a large language model, primarily GPT, to generate menu text based on a prompt styled according to the desired restaurant vibe.

The app’s key design features include [1]:
    • One [OpenAI’s] GPT API call to generate the bulk menu content, followed by
    • A loop on the «Flux» image model for generating and rendering images for each menu item,
    • No external dependencies beyond calling the GPT API, keeping the app fast and accessible,
    • Intelligently consistent pricing and dish categorization,
    • Styling and structure that reflect the cultural and culinary style being simulated.

This tight focus on a single domain - restaurant menus - enables the system to leverage AI’s generative power without overcomplication or heavy infrastructure, although allowing it to excel at expressing the specific style of different culinary types. The whole app functions more as an elegant interface for a targeted generative prompt than as a sprawling multi-service web application [2].

# The Mind Behind the Menu: Andrej Karpathy
To understand MenuGen, it helps to understand its creator. Andrej Karpathy has established himself as one of the most influential figures in modern artificial intelligence. Born in Slovakia, he completed his undergraduate studies in computer science at the University of British Columbia, and has woven a career trajectory that reads like a roadmap of AI's recent evolution.

After completing his Ph.D. at Stanford University, where he studied under the mentorship of AI pioneer Fei-Fei Li, Karpathy worked at OpenAI during its early days. There, he conducted groundbreaking research on deep learning, particularly in the realms of neural networks for computer vision and natural language processing. His blog posts and tutorials during that period helped democratize complex AI concepts, making them accessible to a wider audience of developers and researchers.

In 2017, Karpathy joined Tesla as Director of AI and was instrumental in shaping Tesla’s Autopilot vision and self-driving systems through sophisticated neural network architectures.
 
After leaving Tesla in 2022, Andrej returned briefly to OpenAI before focusing on independent projects and education. Throughout his career, he's maintained an unusual ability to bridge the gap between cutting-edge AI research and practical applications, often communicating complex ideas with clarity and enthusiasm,making him a highly popular figure among both technical and non-technical audiences.

MenuGen represents a good example of Karpathy's approach to AI: technically sound yet accessible, practical yet whimsical, and always focused on creating experiences that feel meaningful to humans[3].

# Vibe Coding: A New Development Paradigm
On February 2nd 2025, Andrej Karpathy shared his coding vision in a tweet that quickly became a defining moment for a new development approach, coining his process as "vibe coding." This expression quickly boosted the imagination of the AI and tech communities worldwide. It consists in producing code to create and deliver a specific vibe or experience rather than focusing on unnecessary complexity or unbounded feature sets, and in using for that purpose the LLM’s and autonomous agents that craft code for us, guided by our continuous flow of written or voiced plain English desires, instead of by precisely tailored coding prompts. 

His exact words were:

 https://img.leopedia.io/DQmT6tM2raodq6GKMRhrFVTEmrmYLHRMAYi2zw5JTgwivpe/The%20%22vibe%20coding%22%20OG%20tweet.png

This short statement resonated deeply in an era where AI development is often obsessed with scale, complexity, and technical benchmarks. Andrej explains that vibe coding involves creating applications where we prioritize the "feel" of the output over technical orthodoxy or feature completeness, also emphasizing minimalism, elegance, and intentionality on the user experience side of things. It also aligns very well with the ongoing evolution the author has described as “Software 3.0,” where AI-generated content coexists alongside traditional software layers to provide creativity and contextual intelligence[4].

The concept of vibe coding has quickly spread, inspiring numerous projects that follow similar principles. In the developers community, "vibe coding" has become shorthand for a method that values the human experience of AI outputs, showing that capturing “vibe” can be as important as technical accuracy. At the same time, MenuGen’s straightforward, browser-based and dependency-free design is one example among many of how practical AI tools can empower everyday users, not just specialized engineers.


**References:**
[1] https://karpathy.bearblog.dev/vibe-coding-menugen/
[2] https://www.latent.space/p/s3
[3] https://www.youtube.com/live/LCEmiRjPEtQ
[4] https://inferencebysequoia.substack.com/p/andrej-karpathys-software-30-software



Posted Using [INLEO](https://inleo.io/@ijatz/from-vibe-coding-to-vibe-cooking-exploring-andrej-karpathys-menugen-text2menu-images-app-bcj)