healthALT

You are what you eat

Comment

Inspiration

The inspiration behind our project is "backend". While we were brainstorming potential ideas to work on, one of my teammates mentioned, "backend" and then i thought we should treat the "backend" of health. The core determinant of health, which is diet and Nutrition. Because really, we are what we eat and what's a better way to solve health problems if not start from the roots.

What it does

Our software simply suggests healthy alternatives of food. It also provide recipes, ingredients for the suggested food. It doesn't stop there, it links consumer to vendors that sells the ingredients or the ready made food itself depending on the their choice.

How we built it

We started by getting the look and feel of our UI in place. Then proceeded to crafting the best possible prompts with the AI model (Claude-3-Opus-20240229) in the early hours of the hackathon. Next, we implemented the UI and after thorough testing (and iterations), we moved on to building the API from which we access the Anthropic model. We concluded by integrating this API with the frontend, and further testing followed

Challenges we ran into

  • The first challenge was identifying which model to use. The best available models aren't free but our curiosity led us to getting free credits.
  • The most interesting part of the hacking was the prompt engineering as we had some real 'fun' getting the AI model to respond in the best possible format for our app to consume (which we eventually figured out).
  • At the end of the day, there was the Vercel deployment issue where we kept hitting different errors (as we're serving a static frontend build from the serverless function). The last error we got was that of __dirname not being defined, a very usual one (the variable is expected to be builtin to the node.js environment but then this StackOverflow answer saved the day:

Accomplishments that we're proud of

  • The last-minute thrill went things went south with unknown bugs due to how we were handling the response from the model. It did feel like we were going to remain stuck at that point, but we made headway at the end. We had a really good prompt but we went a bit overboard with it when we avoided transforming this data too (to get past the bug). But we lost the fact that the previous responses were of a constant and unique format while the AI got even more confused with the updated prompts

We reverted to the old (and stable) prompt and took the time to look into the data transformation (just some objects/array manipulation) and that paid off - and massive thanks to Typescript too xD!!!

What we learned

The technical aspects aside, since the earlier paragraphs have been heavy on them. We've learnt so much in the past few hours, before and during the build, we did extensive research to make sure our solution is needed in the market. And to our surprise, the World needs healthALT. We couldn't believe how much of the little things we neglect concerning our food affect our health. For example;we found out that at least 200,000 Nigerians die annually due to food allergy. But the most interesting thing we learnt was that there is always an healthy alternative for every junks.

What's next for healthALT

We are looking to train an AI model specifically for providing healthy alternatives to food tailored to every consumer personal needs, including large datasets of users' local recipes and ingredients. And eventually, put it out there in the market because we believe we've built a solution the world needs.

Share this project:

Updates