Skip to content

ksek87/sniff_ai

Repository files navigation

Sniff AI

Sniff AI is an AI-powered fragrance creation platform that translates poetic and descriptive text into custom fragrance compositions. This project combines generative AI with a curated fragrance database to inspire and generate scent profiles, showcasing both Product Management and AI/ML engineering capabilities.


Table of Contents


Project Overview

Sniff AI generates custom fragrance compositions based on user-provided descriptions, ranging from simple phrases to poetic imagery. The platform leverages a dataset of existing fragrances for model training and reference, creating unique blends that align with user input.

Check out my Product Management documentation (roadmaps, features, etc) in product-management

Analysis of Fragrances Dataset can be found here


Key Features

  • Text-to-Fragrance Generation: Transform descriptive language into a unique fragrance profile.
  • Fragrance Database: Access to a searchable database of existing fragrances with detailed fragrance notes, scent families, and additional information.
  • Customizable Interface: User-friendly UI for inputting descriptions, selecting preferred notes, and viewing generated fragrance compositions.
  • Feedback System: A mechanism for users to rate generated fragrances, aiding in model refinement.

Tech Stack

  • Backend: Python, Flask or Django
  • Frontend: React or Vue.js
  • Database: PostgreSQL with Elasticsearch for fast search
  • Modeling: Generative AI (e.g., GPT or custom transformer models) and NLP

Project Phases & Progress

Phase 1: Research and Data Collection

  • Gather fragrance data from online databases and public sources
  • Curate dataset of fragrance notes, descriptions, and scent concepts
  • Compile paired data of poetic descriptions and fragrance compositions for initial training

Dataset was built in this project Fragrances

Phase 2: Initial Model Development

  • Fine-tune language model (GPT, BERT, or similar) on paired text-fragrance data
  • Train fragrance composition model to create blends from semantic interpretations of descriptions
  • Evaluate model outputs to ensure relevance and accuracy using similarity scores

Phase 3: Building the Database

  • Set up relational database (PostgreSQL) for storing fragrance data
  • Integrate Elasticsearch for efficient fragrance search
  • Populate database with curated fragrance data

Phase 4: User Interface and Experience

  • Design user interface mockups for text input and fragrance output display
  • Implement frontend in React or Vue.js
  • Integrate model with UI for real-time or asynchronous fragrance generation

Phase 5: Model Refinement

  • Establish user feedback loops allowing users to rate or tweak generated fragrances
  • Integrate feedback mechanism to refine model outputs based on user interactions
  • Iterate on model improvements based on feedback and testing

Installation

Instructions for setting up the project locally will go here.


Usage

Instructions for using Sniff AI will go here, including how to input descriptions, select fragrance notes, and view outputs.


Future Enhancements

  • Additional Input Customization: Allow users to select fragrance intensity, season, or occasion.
  • Improved Fragrance Matching: Refine similarity scoring for more precise fragrance outputs.
  • User Accounts and Saved Fragrances: Enable users to save generated compositions to their profiles.

Feel free to contribute, suggest features, or reach out with feedback. This project is in progress, and any input is appreciated!