• 

• 

  Interface

• 

• 

Inspiration

As the world becomes more digital, accessibility challenges persist, creating barriers for individuals with disabilities—ranging from visual impairments to neurodivergence and motor disabilities. Our solution addresses these challenges by integrating semantic search, AI-driven insights, and accessibility-first principles to revolutionize how users interact online.

By leveraging MongoDB Atlas and AWS services, we demonstrate how combining vector databases with generative AI can dramatically improve usability for people with disabilities. This project offers a scalable and inclusive approach to creating a more accessible digital landscape, empowering users to navigate and perform tasks effortlessly.

What it does

1. Natural Language Query Search

Users can input conversational queries like, “Show me comfortable shoes for winter.” This feature is tailored for individuals using screen readers or voice input, ensuring an intuitive search experience.

2. Error Tolerance

Misspelled or paraphrased queries are automatically corrected and understood. For example, searching for “jens” will still yield results for “jeans.”

3. Personalized Recommendations

Tailored product suggestions simplify decision-making by analyzing user preferences and context, aiding users with cognitive disabilities.

4 . Accessible Product Descriptions

Semantic AI generates simplified, clear product descriptions (e.g., “makeup for sensitive skin”) or categorizes items for better accessibility.

5 . Enhanced Image Search

Using image recognition, users can describe visual elements in natural language, like “blue dress with floral patterns,” for precise results.

How we built it

Core Technologies Google Gemini AI Utilized for generative AI capabilities, enabling contextual understanding and advanced reasoning to enhance product recommendations.

MongoDB with Vector Search Integrated MongoDB Atlas to store and retrieve embeddings efficiently, ensuring fast and accurate recommendations.

Sentence Embeddings Implemented the all-MiniLM-L6-v2 model for robust natural language understanding and semantic similarity matching, key to providing meaningful recommendations.

Key Features

Prompt-Based Recommendations Users can enter any query (e.g., “hydrating skincare products”), and the platform returns tailored suggestions matching their needs.

Generative AI Support Google Gemini enhances recommendations by reasoning about user intent and generating suggestions beyond direct matches.

Embedding-Based Search Semantic AI ensures accurate and fast responses, mapping queries to the most relevant beauty products or catalogs.

Prompt History A sidebar feature allows users to view up to 5 previous queries and their responses for easy reference, improving continuity and usability.

Challenges we ran into

1. Geographic Coordination Collaborating across multiple time zones was a significant challenge, requiring us to adapt to asynchronous workflows and maximize our limited overlap for meetings and discussions.
2. Learning New Technologies Balancing the learning curve of working with Generative AI, embedding models, and advanced search techniques while managing other responsibilities was demanding but rewarding.

Accomplishments that we're proud of

1. Effective Clustering Engine

Successfully building a clustering engine capable of sorting and classifying products in real-time was a pivotal achievement. It demonstrated the robustness of our approach and validated the potential of our product.

2. Scalable Solution

The results motivated us to envision scaling this solution commercially, positioning it as a valuable enterprise asset. With additional refinement, we believe it can address broader industry needs effectively.

What we learned

Key Learnings Through this project, we sharpened our skills in problem-solving, learned about advanced GenAI tools, enhanced our teamwork capabilities, and explored the full potential of AI-driven products clustering.

What's next for AccessMate

Expanded Product Categories Extend recommendations to cover a broader range of categories, including sustainable and hypoallergenic products, catering to niche preferences.

Multi-Language Support Introduce support for multiple languages, ensuring accessibility for users worldwide, especially non-English speakers.

Real-Time Product Availability Integrate real-time inventory checks with e-commerce platforms, ensuring users receive recommendations for in-stock products.

AR Integration for Virtual Try-Ons Incorporate augmented reality to allow users to virtually try beauty products, such as lipsticks or eyeshadows, directly through the platform.

Community-Driven Insights Enable users to share feedback and experiences, creating a community-powered ecosystem for personalized recommendations.

Built With

• geminiapi
• genai
• huggingface
• langchain
• mongodb
• pymongo
• python
• sentence-transformers
• streamlit