Show HN: Afiyah – Snap, Understand Ingredients, Live Clean

1 year ago (a3l17lcdgz1ezx-7860.proxy.runpod.net)

Hi everyone!

I built an app, to make clean and healthy living easier for everyone, introducing Afiyah!

Why Afiyah?

We all want to make healthier choices, but let's face it, understanding ingredient lists is like reading a foreign language. Long chemical names, hidden additives, and tiny text make it frustrating, especially for those with allergies or dietary restrictions.

Existing barcode-scanning apps (Yuka, Sift Food Labels and Processed) don’t always help. They struggle with local, artisanal, or international products without barcodes. And a few that do scan ingredient lists (Ingredio and Trash Panda) often fail to identify ingredients and do not explain why something might be harmful.

That’s why I built Afiyah (Arabic for “health” and “well-being”), a tool that simplifies ingredient analysis in one snap.

What makes it different?

- No more manual research, just snap a photo of the ingredient list.

- Works for products without barcodes, perfect for local finds.

- Gives explanations, not just raw data, about harmful ingredients.

- Saves time and empowers you to make informed decisions effortlessly.

How it works?

Afiyah is powered by a fine-tuned Llama 3.2 Vision 11B multimodal large language model (MLLM), on a dataset I created to extract text from the image without user-driven feature engineering like cropping, as is the case in OCR-based approaches. The system also does not rely on third-party APIs (like OpenAI), enabling:

1. Flexibility: Fine-tuned with PEFT and LoRA for user-specific needs.

2. Continuous improvement: Full control over model updates and optimizations.

3. Cost efficiency: No expensive external API calls.

4. Transparency: No black-box decision-making, Afiyah keeps things clear.

Tech stack: Python, PyTorch, Unsloth, FastAPI, Docker, Gradio, Pydantic Logfire, RunPod.

What's next?

- Gather user feedback for improvements.

- Expand the harmful ingredient database from 150+ to tens of thousands.

- Enhance MLLM accuracy and efficiency.

- Launch a mobile app for on-the-go use!

For more details and comparisons, read: https://hasibzunair.medium.com/afiyah-snap-understand-ingred...

Do consider trying it out and make sure to leave a feedback, either on the app or in the comments. Let's win clean living together!