Workflow

Inspiration

Labels are everywhere, and ingredients (especially in skincare, cosmetics, and medicine!) are overwhelming. With thousands of obscure ingredients and conflicting advice, users need a neutral, science-backed tool to decode ingredient labels. LabelLens was born to democratize ingredient literacy!

What it does

LabelLens empowers its users by allowing for simple, un-biased, and convenient way to look up ingredients. Users can either take a picture of an ingredient list to get a quick summary of each ingredient, look up a specific product's information and ingredients, or look up a specific ingredient themselves.

Market Research

Most competitors fail users in two ways:

Bias for Profit: Many apps quietly partner with brands, skewing results or hiding unfavorable data.
Stale Data: Manual updates can’t keep pace with new ingredients (e.g., vegan alternatives, novel actives).

LabelLens combats both with

RAG + Google Gemini: Our AI cross-references real-time research and vetted databases, avoiding human curation bias.
Zero Brand Partnerships: We prioritize user trust over monetization, ensuring analyses are never influenced by third parties.

How we built it

React.js with Tailwind for the frontend
Node.js for the backend
Google Gemini and GCP (Docker, Kubernetes) for the AI, Weaviate for the vector database
Data was collected from various reputable sources, such as the FDA.

Challenges we ran into

Finding and acquiring appropriate (non-biased) datasets of ingredient side-effects for retrieval augmented generation (RAG).
Deployment of a self-hosted vector database on the cloud for our RAG system and populating it with relevant data
no food

Accomplishments that we're proud of

Hosting the website using Google Cloud Platform
We're proud to have learned to combine custom RAG algorithms with Google Gemini

What We Learned

Cursor is a good AI

What's next for LabelLens

Comparing two items (e.g. If two items are )
Providing alternatives (e.g., I like a product, but it contains an ingredient I do not like. Is there an alternative at the same level(?) but without the ingredient?)
Fully integrate the RAG system