跪拜 Guibai
← All articles
Artificial Intelligence · Frontend · Agent

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

By 雨夜寻晴天 ·
Read original on juejin.cn ↗ Google Translate ↗ Alt translation

Meal-planning AI that only outputs dish names creates more work than it saves. Yanhuo Shijian closes the gap between a generated menu and an actionable plan by linking recipes, nutrition data, and a consolidated shopping list, then handling the failure modes that cause users to abandon a tool.

Summary

Yanhuo Shijian tackles the daily friction of answering "what's for dinner" by generating full meal plans that respect real-world constraints: number of people, dietary restrictions, health goals, budget, and available cooking time. Unlike a simple AI prompt that spits out dish names, it produces an executable plan that links each dish to its recipe, nutritional estimate, and required ingredients, then aggregates everything into a single shopping list where purchased items can be checked off.

The tool also handles the unglamorous edge cases that make or break a product. It recovers from model timeouts, preserves in-progress menus across page refreshes, degrades gracefully when the database is unreachable, and adapts its layout for mobile screens. The source code remains private while the product iterates rapidly, but a public GitHub repository tracks the roadmap, changelog, and bug reports.

A live version is available at yanhuofood.com, and the creator is actively collecting real-world usage scenarios — cooking for elderly family members, squeezing meals into 20-minute windows, or planning a single weekly grocery run — to guide further development.

Takeaways
Yanhuo Shijian generates weekly menus constrained by cuisine, dietary restrictions, health goals, number of people, meals per day, budget, and cooking time.
Each dish links to a recipe, ingredient list, nutritional estimate, and a cooking video or Douyin search entry.
Swapping a single dish updates the aggregated shopping list incrementally rather than regenerating it from scratch.
Purchased and already-at-home items can be marked off the shopping list directly in the interface.
The tool handles model timeouts, partial data returns, database unavailability, page refreshes during generation, and mobile layout adaptation.
Source code is not yet public; a GitHub repository serves as a public roadmap, changelog, and issue tracker.
The creator is collecting specific real-world scenarios — tight cooking windows, multi-generational households, budget caps, once-a-week shopping — to prioritize features.
Conclusions

Most AI meal planners stop at generating dish names, which leaves the user to manually cross-reference recipes, compile a shopping list, and reconcile dietary conflicts — the actual labor the tool was supposed to eliminate.

Treating menu generation as the start of a workflow rather than the end product shifts the value from "AI wrote something" to "AI produced a plan I can execute without further effort."

The decision to keep source code private during rapid iteration is a practical acknowledgment that maintaining a public codebase and a fast-moving product are competing priorities, and doing both poorly helps no one.

Collecting granular, unglamorous usage scenarios — like cooking for elderly family members or fitting meals into 20-minute windows — is a stronger signal for product direction than generic feature requests.

From the discussion

The conversation centers on a feature request for takeout recommendations. The original poster wants the app to eliminate decision fatigue by simply telling them what to order. The developer confirms the idea is on the roadmap but asks for clarification on the scope. A reply argues that full system integration is overkill and that a simple suggestion feature, despite the challenge of scraping location-specific data, would suffice. A separate request asks for per-meal dish quantity settings.

A takeout recommendation feature is requested to solve daily decision paralysis about what to eat.
The developer is considering the feature but is undecided between a full integration with delivery systems or a simpler suggestion-only approach.
Full takeout system integration is seen as too complex; a simple suggestion feature is preferred, though it requires scraping city-specific platform data.
Meal planning should allow different numbers of dishes per meal, as breakfast typically requires fewer items than lunch or dinner.
Featured comments
Ang3rliu 1 likes

OP, I order takeout every day. Could you add a feature to recommend takeout? It's not that I have many mouths to feed, I just never know what to eat for each meal, and I don't really crave anything specific. I just don't want to agonize over it. If it could automatically arrange what takeout to order for me, I'd just order and eat, and wouldn't have to struggle with deciding what to eat anymore.

雨夜寻晴天

In the planning stage. Takeout — we have this idea, but are you thinking of integrating with a takeout system, or just suggesting what to eat? It's in the follow-up roadmap.

Ang3rliu  → 雨夜寻晴天

Wouldn't integrating with a takeout system be a bit too much trouble? I think just telling me what to eat would be enough. The only thing is you'd need to scrape data from takeout platforms, since it kind of depends on the city. [wronged]

See top comments, translated →
Source: juejin.cn ↗ Google Translate ↗ Backup ↗