Chaat King

• 

  Name and Goal

• 

  Tools for calling Backend APIs

• 

  All Examples

• 

  Prompt

• 

  Example Deep Dive

Inspiration

My inspiration for Chaat King came from the everyday challenge of dealing with leftovers. I wanted to create a solution that not only reduces food waste but also sparks creativity in the kitchen. The idea was to make meal planning more sustainable and enjoyable, turning what might seem like uninspiring leftovers into exciting new dishes.

What it does

Chaat King is a chatbot that helps users transform their leftovers into delicious recipes. Users can input the ingredients they have, and the bot suggests creative recipes, provides a downloadable recipe for social sharing, and tracks the nutritional value of the meal. It's designed to make meal planning easier, reduce food waste, and promote healthy eating.

How we built it

I built Chaat King using Vertex AI Agents. It involved adding and tuning the prompt, vibe checks and creating examples that the AI can refer to. I also used another API to create images and videos, which can be integrated using tools. I also explored adding nutritional value of various food items, and this can be done by using data stored in buckets or tables.

Challenges we ran into

UI is a little bit clunky, here's a couple of issues I found:

1. UI for creating a data store is stuck at a particular place I can't click Continue, and it also says "Some form fields are incorrect". None of the fields give me insight about which one is wrong. My 2 cents as a developer here is that some state management in react is probably broke: Let's say I have an issue in the URL patterns field, and I fix it, looks like that step only updates the field state, but not the side bar state.
2. Creating an agent runs into this issue, not a good look with early adopters, but I am sure someone will get it sorted soon :)
   
3. Can't say if this is a bug or a feature, but: the LLM gets stuck on a repeat loop -- I've seen this with gemma non-finetuned, but with gemini that's a surprise. My guess is it needs more finetuning, or a guardrail against this in inference.

Accomplishments that we're proud of

I am proud of exploring the UI, all the options, and creating a demo chatbot that not only addresses the issue of food waste but also encourages healthy eating habits.

What we learned

Throughout the development of Chaat King, I learned about the complexities of generative agents and the importance of accurate data, right prompt, and example. I also learned that accurate data can be provided using tools like data tables, or API calls for nutritional analysis. I also gained insights into user behavior and preferences when it comes to meal planning, leftover management, and levels of engagement due to introduction of multi-media.

What's next for Chaat King

Moving forward, I plan to expand Chaat King's recipe database to include a wider variety of cuisines and dietary preferences. I also aim to integrate a feature that allows users to save their favorite recipes and track their meal history. Additionally, I am exploring partnerships with grocery delivery services to provide users with the option to order missing ingredients directly through the app.

Built With

• vertexai