Hello Chef

Your AI sous chef!

Comment

Inspiration

I've been very interested in learning to cook every since I got married. I've always just kinda winged it, but then came chatgpt. It was awesome to be able to get a recipe, substitute things and learn interesting ways to cook to make delicious meal. An example of this is when I was cooking 'the worlds most fluffiest pancakes' and realized i didn't have butter milk. Chatty-G said I could just swap in regular milk that I added a tbsp of an acid like vinegar or lemon juice to, to get the same effect. I would've never learned that from a traditional recipe or cooking site. Then one day on a drive (after learning about function calling to get structured data) I was thinking about how tedious it is to enter pantry items into a list to then see what you can make from them, so I had an idea. That night when I got home, I spun up a web app where I used voice recognition to talk with chatgpt and function calling to convert a spoken list of items in the pantry into a structured json out put that I could store in a db! It was awesome!

That was the start, and after a couple year of ideation, this is where we landed! Hello Chef - Your AI Sous Chef A companion to help you in the kitchen from anything from ideas, to plans, to guides, to shopping! And you get to do it in community!

What it does

Hello Chef is a mobile cooking app built around Chefie, an AI sous chef.

You can talk to Chefie the way you would talk to someone standing in the kitchen with you: "I have chicken, spinach, and half an hour," "make this dairy-free," or "what can I substitute for buttermilk?" Chefie can answer the question directly or use tools inside the app to find, create, save, favorite, and refine recipes.

There are three ways to add a recipe: talk it through with Chefie, use a structured recipe prompt, or enter one manually. Recipes can be searched, edited, saved, favorited, shared, and organized into cookbooks.

When it is time to cook, Hello Chef turns a recipe into a step-by-step guide with serving adjustments, timers, and voice-friendly help. Chefie knows which step you are on, so you can ask a question without losing your place.

The app also connects recipes to the rest of the week. You can keep a pantry, ask Chefie to make a meal-plan draft, review or swap individual meals before saving anything, and turn a meal plan or cookbook into a grouped shopping list. Dietary preferences and saved allergies are carried into generation and checked again by deterministic server rules.

Hello Chef also has public chef profiles, attributed recipe adaptations, follows, a following feed, and public or private cookbooks. If someone makes their own version of another cook's recipe, the original stays credited.

How we built it

The app is built with Expo and React Native, with Expo Router handling the native navigation. The backend is a Next.js and TypeScript service with regular HTTP routes for reads and short actions, plus WebSockets for conversations and longer-running recipe generation.

Chefie uses the OpenAI Responses API with a grouped set of typed tools. Those tools cover recipe search and creation, recipe changes, guided cooking, pantry and shopping-list updates, and meal-plan drafts. Every tool call is parsed and validated on the server before it can touch user data.

Recipe generation streams progress back to the phone so the interface can show what is happening instead of sitting on a spinner. Recipe images are generated in the background and delivered through an Inngest job. PostgreSQL stores recipes, preferences, meal plans, shopping lists, conversations, and recipe history, with Prisma providing the data layer.

I actually built some of the app manually in Nest.js for the backend and expo for the mobile app. But recently I decided to pick it back up with 5.6 sol. I put chatgpt in voice mode and just dumped everything I wanted this app to be both in technical and non-technical jargon and let it organize everything into a giant list. I have 5.6 sol take that list and break it down into a checklist of implementation steps in extreme detail. It had over 1000+ items. Including items that allowed it to control my browser to make sure my clerk, apple, google, vercel were all updated with the proper items.

After a little testing and iteration, this is what we have. It still has plenty of room for improvement, but it works!

Challenges we ran into

The hardest part was making Chefie useful without giving it vague, unlimited control. Cooking is conversational, but saving a recipe, changing a meal plan, or deleting something is not. I ended up drawing a clear boundary: the model can decide which culinary tool is appropriate, but ownership checks, schemas, quotas, idempotency, and destructive-action confirmation live in normal application code.

Meal planning was another deceptively hard problem. A model can suggest seven dinners, but it should not be trusted to do ingredient arithmetic or silently overwrite a user's Tuesday breakfast. The final flow asks the model for a structured draft, checks every slot and restriction outside the model, lets the user review and edit it, and only writes the accepted plan.

Streaming was also tricky. A single request might involve understanding the prompt, searching existing recipes, creating one or more new recipes, validating them, saving them, and queueing images. Making that feel responsive on a phone required a shared event protocol and careful recovery when a connection or model call timed out.

Finally, I spent a lot of time on the unglamorous parts that turn a demo into something I would actually ship: privacy controls, recipe ownership, allergy safeguards, account deletion, rate limits, offline grocery checkoffs, notifications, subscription verification, and testing the same contracts across the app and backend.

Accomplishments that we're proud of

The part I'm proudest of is that Hello Chef is no longer a collection of mockups and half-finished experiments. It is an app I can actually use: Chefie streams responses in real time, recipes flow into guided cooking and shopping lists, and the social pieces work without losing attribution. I also managed to build the less visible product work—privacy controls, allergy checks, ownership rules, and account deletion—instead of leaving it for later. Codex helped me move fast enough to tackle that whole surface area during Build Week, but every feature still had to pass the same test: would I trust this in my own kitchen?

What we learned

Building this changed how I think about AI products. The model was rarely the hard part; the hard part was deciding exactly what it should be allowed to do, validating the result, and keeping the interface honest about what was happening.

Typed tools beat one giant prompt. A separate draft-and-confirm step made meal planning safer, and streaming small progress events made long recipe jobs feel much better on a phone.

I also learned to keep design work distinct from implementation. I used Codex heavily for coding and audits, but the voice, interaction patterns, and visual decisions still needed direct, opinionated attention. The biggest practical lesson was to spend time on failure paths early: timeouts, duplicate requests, image jobs that finish later, and half-completed writes are all normal in an AI app.

What's next for Hello Chef

The immediate next step is getting the iOS release through App Store review and putting the app in front of home cooks who were not involved in building it. I want to learn where Chefie genuinely saves time and where the experience still asks too much from the user.

After that, I want to improve recipe quality from real cooking feedback, make pantry updates even faster, and keep tightening voice interaction for messy, hands-full cooking. Longer term, I would like to add household collaboration and grocery fulfillment once the right integrations are available.

The goal stays simple: make the distance between "what should I cook?" and dinner on the table feel a lot shorter.

Built With

Share this project:

Updates

Submission history