AppCookBook - pitched but team was not assembled

Inspiration

Its like Spotify for Apps. For non developers and developers alike. Making and mixing apps one ingredient at a time.

In March 2023 OpenAI announced Code Interpreter. Feature for ChatGPT that based on natural language input writes python scripts that solve user problems. It has its limitations but nothing better happened since for non developers. Tools like Github Copilot were released. But they are developer assistants. Many other tools were released that claim to be Software Engineer Agents. GPT Enginer, Smol Dev, GPT Pilot. They write apps from scratch. And need to deal with enormous breath of contextual uncertainty that causes a lot of errors. Operating systems, versions, environment setup, versions of all the software... In my experience of using them myself or especially giving them to others... They encounter errors all the time and those errors can not be overcome without experienced developer. So they write from scratch and encounter same errors all the time. This makes them slow, expensive and outright impossible to use for non developers.

This is where AppCookBook comes in.

So what we want to target is a long tail of demand for software. Entrepreneurs who need to build and MVP. Developers who have no idea to build all small app requests that come their way and don't even have time to write all apps they want themselves. But true target are people who are far from IT world and yet still have need for silly personal apps for their daily challenges.

Our solution secret sauce? Two things. Opinionated predictable environment that removes a lot of uncertainty and possibility of errors. And code reuse.

When user comes to AppCookBook and describes what he wants following things happen:

1. AI searches in already made, used and tested data base of app recipes. If it finds similar app it can adapt. Good! It adapts it for the user almost without writing code. Fast, cheap and with almost 0% chances for errors.
2. If it does not find a ready recipe it crafts a list of ingredients it would need and searches for them. If it finds them it glues them together to serve the user needs. Still way cheaper and faster then writing from scratch. And less opportunity for errors. And if it works? It get's saved.
3. If ingredients are missing? Well then it can write them from scratch. But it still small single purpose block of code. AI can write those today. It may still not work out. But that is where developers come in. Platform will surface failing ingredients and how many people want them. And with time platform will grow so that less and less cases are not covered.

Monetisation? Spotify style. Subscriptions are divided between recipes and ingredients that were used by the user. Which puts incentive for developers to developer good ingredients for the platform that will be used by AI a to serve the needs of a lot of users in a lot of niche use cases.

What it does

Current prototype is a node server app. It can be started at your server or desktop machine. It exposes API for custom ChatGPT bots to use terminal. This allows it to manipulate file system, install libraries and apps, write, execute and test code. It was already used to expand and rewrite itself.

But for this hackaton we are focusing on frontend web apps. HTML/CSS/JS. Predictable environment that has no versions and works the same on your phone, desktop or even a TV. We load 3rd party libraries from CDNs and can make a lot of things happen, from Firebase based distributed collaboration apps to vector editors, to HuggingFace transformer.js models running in browsers locally. There are limitations but a lot is possible and useful too.

For a hackaton goal is to test product market fit, finish missing things and start filling it in with recipes and ingredients based on what other teams at hackaton need :) Hackaton participants are first clients ;)

How we built it

Its built with JavaScript/Node/Express/localtunnel and bunch of other libraries. Connected to private CustomGPT and from there GPT4 was helping to build it by editing its own files.

Challenges we ran into

At this hackaton it was harder then expected to assemble the team on the spot so project was not started

Accomplishments that we're proud of

...

What we learned

To be safe come with assembled team as not all hackatons have good matchmaking on spot

What's next for AppCookBook

Plan is to continue exploring idea validity and product market fit outside of hackaton

Built With

• express.js
• javascript
• node.js