Yanhuo Shijian Turns AI Meal Planning Into a Shopping List That Actually Works

Every day at mealtime, one question always pops up right on time:

What should we eat today?

It looks simple, but answering it seriously immediately turns into a string of questions:

• What do you feel like eating?
• How many people are at home?
• Does anyone not eat spicy food?
• Are you trying to cut fat lately?
• What's left in the fridge?
• How much time can you spend cooking today?
• Once the dishes are decided, what exactly do I need to buy?

Eating alone, you can still get by with takeout and "whatever." When there are more people, things are completely different.

Someone wants something light, someone else can't live without chili; one person is counting calories, another is a growing kid. After finally figuring out the menu, you miss two items at the grocery store and only realize you have to go back out when you get home.

I built Yanhuo Shijian (烟火食间) to solve this small but daily hassle.

Product experience address: https://www.yanhuofood.com

Public product repository: https://github.com/you-want/yanhuofood

At first, I thought AI just needed to spit out a few dish names

When I first started this product, I thought it was simple:

Users fill in their preferences, AI gives a weekly menu, done, right?

Only when I really got into it did I realize that generating a few nice-sounding dish names is still a long way from producing a truly executable menu.

For example, AI might arrange a week of non-repeating dishes that look very rich, but there could be these problems:

• Every dish requires buying a completely different set of ingredients.
• A weekday dinner is scheduled with a dish that needs two hours of stewing.
• Dietary restrictions were clearly filled in, but the menu still includes them.
• The calorie count looks healthy, but it's not enough food for a family of three.
• After the menu is generated, you still have to manually copy the grocery list.

That kind of menu is good for a screenshot, not for daily life.

So now, when using Yanhuo Shijian, you can first lay out the actual situation for this time.

Cuisine type, dietary restrictions, health goals, special populations, number of days planned, meals per day, number of people, dishes per meal, budget, and cooking time can all be adjusted.

If you have free time on the weekend, make it a bit richer; if you get home late on a workday, try to arrange quick dishes. Traveling, eating out for work, and weekend meal prepping should also have different planning methods.

Frankly, AI can't just "improvise"; it also needs to listen.

After the menu is generated, things just get started

I increasingly feel that the menu is just the starting point of the entire process.

A truly useful product should continue to answer these questions:

• What ingredients does this dish need?
• How exactly do you make it?
• Roughly how many calories?
• I don't like this dish, can I just swap it out?
• Adding up all the dishes, what exactly do I need to buy?
• Which items are already bought, and which are already at home?

So Yanhuo Shijian now strings together the menu, recipes, nutrition estimates, dish swapping, and the shopping list.

After the menu is generated, you can continue editing. If you don't want a certain dish, you can swap it; if the whole meal isn't right, you can rearrange it.

Open the dish details to view ingredients and steps. Recently, dish images, cooking video links, and an entry to jump to Douyin to search for related cooking methods were added.

Once the menu is confirmed, the system automatically summarizes the shopping list. Items already bought or already at home can be directly marked, no need to manually copy the menu again.

What I want to build is not a tool where "AI sends you a block of text," but a tool that can ultimately accompany you to the market, the kitchen, and the dining table.

Building this product was way more trouble than I expected

If you just watch a demo, AI menu generation looks like pressing a button, waiting a few seconds, and a table appears.

But what really affects the experience are often the situations that don't appear in demos:

• What if the model request suddenly times out?
• What if the returned data is missing a day or a meal?
• If the database is temporarily unreachable, can you still view previously saved menus?
• If the user refreshes the page, does the in-progress generation task just disappear?
• After swapping one dish, can the shopping list only add and remove the corresponding ingredients?
• A table that looks good on a widescreen, will it be all squished together on a phone?

These things aren't "flashy," but they determine whether a product can actually be used.

A lot of my time recently hasn't been spent piling on more AI features, but on handling failure recovery, historical data compatibility, mobile layout, and local data degradation.

Because users don't care what technology is behind it. Users only feel: Why did my menu from last week suddenly disappear? Why can't I tap this button? Why did it get stuck halfway through generating?

By tackling these small problems one by one, the product slowly becomes more reliable.

Why isn't the source code open source for now?

Since the GitHub repository is already public, some might ask: Where is the source code?

Currently, the application source code for Yanhuo Shijian is not open.

The reason is actually not complicated: At this stage, I want to focus my main energy on the product first.

The product is still iterating rapidly; feature boundaries, data structures, and interaction methods are all changing. If I just open-source the source code directly now, I don't want to just throw out a repository under construction and call it "open source complete."

Truly responsible open source requires doing many more things:

• Organizing code and module boundaries.
• Completing installation, deployment, and contribution documentation.
• Handling security configurations and privacy boundaries.
• Maintaining Issues, Reviews, and community communication.
• Being responsible for the time of external contributors.

These all require continuous investment.

If I'm rapidly changing the product on one hand, and spending a lot of time packaging every temporary implementation to look suitable for open source on the other, I worry that in the end, neither the product nor the open-source project will be well maintained.

So my current choice is:

The source code remains private for now; focus on making the product good first.

This doesn't mean it will never be open source in the future.

Once the product boundaries are more stable, I will consider gradually opening up common data formats, utility modules, or other parts suitable for independent maintenance, and I won't rule out re-evaluating the possibility of a broader open source later.

But for now, I'm not setting a timeline or making promises in advance. Make the product worth using first, then discuss how to open it up.

If the source code isn't open, why build a public GitHub repository?

Although the source code isn't public for now, there's no need to hide the product's progress.

This public repository will continuously contain these things:

• Product introduction and real screenshots
• Update logs
• Product roadmap
• Bug report entry
• Feature suggestion entry
• Changes to future open-source plans

Repository address: https://github.com/you-want/yanhuofood

If you find this direction somewhat interesting, you can give it a Star.

A Star isn't just a number to me; it at least shows that it turns out I'm really not the only one being tossed around daily by the questions of "what to eat, what to buy, how to cook."

Finally

Yanhuo Shijian isn't a finished product yet; there are still many areas that need polishing.

But I hope it can ultimately accomplish one very simple thing:

Let ordinary people spend a little less time agonizing over what to eat, and a little more time truly eating well.

If you're willing, you can try it directly:

• Product experience: https://www.yanhuofood.com
• GitHub repository: https://github.com/you-want/yanhuofood
• Issues and suggestions: https://github.com/you-want/yanhuofood/issues

If you encounter problems during actual use, you're also welcome to tell me the specific scenario.

For example, having elderly and children at home, only having twenty minutes to cook each day, wanting to control the budget, wanting to buy groceries for five days in one trip, or simply not wanting to eat the same few dishes every day.

These real, specific, even somewhat trivial problems are exactly what I want to continue solving next.