Description:
How I designed the first agentic AI beauty assistant for girls that remembers your skin, matches your skin tone, undertone and personalises every recommendation just for you.
The Problem Nobody Is Talking About:
You pick up your phone, scroll through the various shades of a makeup product, say, a foundation, to find the one that matches you perfectly, and then place your order. Four painstaking days pass, and when you apply it, you are disappointed because it does not give you the perfect shade, it might be too grey, or it may be orange, or it simply doesn’t suit you.
This is not an unusual experience.
Recent data states that about 40% of foundation sales in India end up being returned due to the shade mismatch. Moreover, what about all those millions of people living in non-metro cities where there are no physical beauty stores? There is absolutely no guarantee with regard to what shade will work on your skin.
The solutions that are currently in place are not that effective. Your undertone, skin type, past reactions, and the fact that your skin in the summer differs greatly from your skin in winter; these all are all ignored by camera-based shade-matching apps, which provide you with a generic result. As soon as you exit the app, they forget about you. They give everyone the same product recommendations. We refer to it as the "wrong shade syndrome." And it hasn't been resolved for far too long.
Introducing GlowMind:
Every skin needs its ideal shade, and everyone shines in their unique way. GlowMind is not merely a chatbot, nor is it any kind of shading filter, nor is it some regular beauty advice software. GlowMind is an autonomous, agentic AI-driven system that safeguards your skin from any dangerous mixtures, keeps your skin data in memory, thinks twice before speaking, and learns with each dialogue.
The Concern In Detail:
It's crucial to comprehend the breadth of GlowMind's problems before learning what it does.
There is such a thing as Wrong Shade Syndrome. Under different lighting, camera settings, and screen calibrations, foundation, blush, and concealer appear significantly different. Under office lighting, a warm beige that appears flawless on your phone screen may appear ashy and incorrect on your skin. Camera-based matching does not capture the true nature of your skin; it only records what the camera perceives.
The silent killer of beauty e-commerce is a lack of personalisation. Current applications rely on generic databases and one-time camera scans. The last time you used a fragrance-heavy foundation, you were unaware that you had breakouts. They are unaware that benzoyl peroxide dries out skin while niacinamide works wonders. There is such a thing as Wrong Shade Syndrome. Under different lighting, camera settings, and screen calibrations, foundation, blush, and concealer appear significantly different. Under office lighting, a warm beige that appears flawless on your phone screen may appear ashy and incorrect on your skin. Camera-based matching does not capture the true nature of your skin; it only records what the camera perceives.
Personalisation that fails to personalise remains the silent killer in e-commerce of beauty products. The technology in use today depends upon universal databases and single-use camera captures. It ignores the fact that the last time you used a foundation with heavy fragrances, your skin had an adverse reaction. Understanding the extent of the challenge that GlowMind addresses will help you comprehend what it really does. There is a term known as Wrong Shade Syndrome.
What Makes GlowMind Different: The Four Novelties
Here’s what GlowMind can do that no other beauty app can do.
Permanent Memory
While most beauty apps ignore your existence when you close their app, GlowMind remembers everything. Using a database structure in MongoDB Atlas, GlowMind saves your skin colour, undertone, skin type, issues, used products, reactions, budget, and regimen. Each Streamlit session starts with loading your full history, which only adds up to your profile and makes you even more recognisable. Your skin profile gets more and more sophisticated and composed with every session.
Reasoning Chain for ReAct
Almost every machine learning model works like a black box. You query it, and then it gives you an answer without a human-reasonable explanation. Not GlowMind! By implementing the ReAct methodology, GlowMind reasons with you: it gives a clear description and reason for each step it takes while coming up with recommendations and provides explanations for every recommendation made.
Identification of Ingredient Conflicts
This is the most unique feature in any beauty application and GlowMind's most protective feature. Vitamin C and retinol should never be taken together because they irritate one another and lessen each other's benefits. When combined, retinol and AHA seriously harm the skin's barrier. Retinol is totally deactivated by benzoyl peroxide. But unfortunately, the majority of people are unaware of this. This is something that most beauty apps don't tell you. Instead of alerting you after the damage is done, GlowMind proactively warns you before you combine incompatible actives.
Seasonal Skin Drift
This is a fact that your skin does not look the same in January as it does in July. Throughout the year, your skin type and tone are influenced by temperature, humidity, sun exposure, and various lifestyle changes. GlowMind automatically asks you to update your profile if it is older than six months. Your skin's natural seasonal changes are not taken into account by any other beauty product.
Pipeline:
GlowMind uses a six-stage pipeline to process each query. These are as follows-
User input, memory recall, LLM reasoning, tool execution, live search, and personalised response.
Your input, which is a questionnaire taken during onboarding, is where the journey starts. GlowMind stores the user profile in MongoDB with a unique user ID and loads the whole profile from there.
During the reasoning phase, the Python functions according to the keywords in the query.
Three tasks are carried out by the tool's execution: BERT sentiment analysis on actual product reviews, ingredient conflict checking against a curated database of hazardous active pairs, and shade matching against 9000 actual Sephora products
Every recommendation comes with actual purchasing information because Tavily's live search retrieves current prices and availability from the internet.
Lastly, GlowMind provides a friendly, well-reasoned, detailed recommendation that stays within your budget, explains each option, and alerts you to any conflicts.
Technology Stack:
Groq API: Free inference platform that runs Meta's Llama models at extremely high speed. We have used llama-3.3-70b-versatile to handle complex beauty queries and llama-3.1-8b-instant for fast responses.
Hugging Face BERT: for reading product descriptions and reviews in the Sephora dataset, and gives sentiment scores to prevent GlowMind from recommending a product with low reviews.(Avg. score taken 3 to reduce time)
Kaggle Sephora Dataset: Contains 9000+ real beauty products with shade codes, ingredient lists, ratings, reviews, and their prices. This is used as the database for RAG retrieval
Tavily: This gives the ability to search the internet for current product prices and availability for Indian websites such as Nykaa and Myntra.
MongoDB: This is a lightweight database that stores every user's skin profile permanently across sessions.
Streamlit: This is an interface that turns all the Python code into a beautiful, shareable web app.
User Experience :
Starting GlowMind is actually very different from starting any beauty app. There won't be any spamming ads or generic tips which will fill up your phone's screen.
Just start typing as you talk to a close friend. Are there any foundations that work for under 800? Should I use something against dark spots and hyperpigmentation?
GlowMind uses your data and available information to give personalised step-by-step answers without breaking your budget.
Future Work:
The next release will also include computer vision for shade detection, which will perform hex-code matching; the user's skin tone hex code will be perfectly matched with the product's shade. A mobile app will be convenient to use for users, providing personal beauty intelligence for users in every city. The Indian brand list will also be expanded to include more brands like Lakme, Sugar, Colorbar, Mamaearth, and many more.
But the goal is the same. Every skin tone has a perfect match. Every person has a unique glow. GlowMind is dedicated to finding it.
About:
GlowMind is conceptualised and being created by Ayushi Shukla as a part of the BCA seminar project. It is a combination of agentic AI, personalisation, and real-world beauty tech.
Built with Groq, LangChain, Hugging Face, Sephora Dataset, Tavily, and Streamlit.
Match Your Magic.
Top comments (0)