Planistry: How It Works

Planistry reimagines wedding vendor discovery by turning visual inspiration into personalized, high-quality matches. In the main case study, I shared how I designed the platform around real couples' needs, from early research to prototyping and testing.

Here, I unpack a conceptual AI model that outlines how the system could work behind the scenes: how it would be trained, how it might make recommendations, and how the experience would stay grounded in human-centered design. I also touch on key considerations around responsible AI and share thoughts on future opportunities as the system evolved.

How the Model Learns

Planistry’s AI model relies on computer vision and vector similarity to interpret inspiration images and surface relevant vendor matches. To make this possible, the model first needs to “understand” what’s in an image, such as style, mood, and visual elements like color palette or floral arrangements.

The model is trained using supervised learning, a method where the algorithm learns from labeled examples. In this case, each training example pairs an image with descriptive labels that tell the model what it’s looking at. This labeled data acts as the foundation to guide the model’s learning.

Training data sets include:

  • Wedding item images (e.g., bouquets) with captions (e.g., “classic red rose bouquet”)

  • Vendor categories (e.g., florist, baker) linked to item types they provide

  • Vendor business names connected to categories (e.g., “Sally’s Flowers” → florist)

  • Vendor names paired with images of their past work

  • Vendor names linked to service locations (e.g., Arlington, VA)

During training, the model converts images and text labels into numerical representations called vectors. These vectors capture the essence or “meaning” of the input in a format the machine can understand. The model then learns to maximize cosine similarity (a measure of how closely aligned two vectors are) to match images with their correct labels. Conversely, it minimizes similarity between unrelated pairs to avoid incorrect matches.

This process allows the model to recognize not only exact matches but also stylistic similarities, enabling it to recommend vendors whose past work visually aligns with a couple’s inspiration, even if the images aren’t identical.

Supervised learning and vector similarity drive Planistry’s image-to-vendor recommendation process.

How the Model Makes Recommendations

Once trained, the model uses the same vector-based approach to interpret user-uploaded inspiration images. It converts each uploaded image into a vector, capturing key visual features and style elements.

Next, the model compares these user image vectors to the vectors representing the vendor portfolio images. It calculates cosine similarity scores to identify vendors whose work visually matches the couple’s inspiration.

To ensure recommendations are practical, the system filters potential matches based on user criteria like location, vendor availability and other user-inputted information, narrowing the list to vendors who not only fit stylistically but are also feasible options for the couple.

Finally, these filtered matches are ranked by similarity score and presented to users as an ordered list of vendor recommendations. This ranking balances AI-driven style matching with real-world constraints, helping couples discover vendors that truly fit their vision and needs.

Sample vendor results display: ordered by match score, filtered by location and user inputs.

Human-Centered Design Choices

Planistry was designed to give users a sense of control and clarity as they navigate vendor discovery:

  • Transparent Matching: Users receive a ranked list of vendor matches with a percentage match score, helping them understand how well each vendor fits their inspiration.

  • User Inputs for Better Accuracy: The model incorporates additional details like wedding location, venue, and budget to filter and improve the relevance of recommendations, ensuring suggestions are both stylistically and practically aligned with the couple’s needs.

  • Interactive Features: Users can tag favorite elements in inspiration images and provide like/dislike feedback on vendor matches. This feedback loop continuously refines the model’s accuracy.

  • Supportive Chatbot: The Planistry Companion chatbot guides users who may be unsure about their vision by using natural language processing (NLP) to talk through their ideas. It also breaks down complex planning tasks and provides personalized encouragement, making the experience feel guided and less overwhelming.

Planistry’s chatbot provides vision clarification, emotional support, and task unpacking via conversational AI.

Responsible and Ethical AI Considerations

  • Mitigating Data Bias: Wedding inspiration and vendor portfolios often reflect dominant cultural aesthetics and popular trends, which can inadvertently bias recommendations. To address this, Planistry’s model is trained on diverse data sets representing a wide range of styles and cultures.

  • Balancing Vendor Exposure: The recommendation algorithm incorporates fairness-aware techniques, such as weighting vendor representation and periodically refreshing vendor rankings, to ensure that emerging vendors have a chance to be discovered. Additionally, the system includes mechanisms to boost visibility for new or smaller vendors who may have fewer portfolio images, preventing them from being overshadowed by established vendors with larger image libraries. This helps create a more equitable marketplace and encourages vendor diversity.

  • Geographic and Vendor Data Constraints: Since vendor availability and styles vary widely by location, Planistry deliberately filters matches to vendors serving the user’s wedding area. While this limits recommendations to known data, it ensures relevance and avoids suggesting unavailable vendors.

  • Enhancing Transparency and User Trust: By displaying clear match percentages and offering the Element Guide that explains style labels, Planistry empowers users to understand why vendors are recommended. This transparency helps prevent blind trust in AI and encourages users to make informed decisions.

  • Privacy and Secure Data Handling: Couples’ personal details, wedding dates, and inspiration images are sensitive information. Planistry follows strict data protection practices, encrypting data in transit and at rest, and limiting access to authorized personnel only. Vendor information is also managed respectfully to maintain confidentiality and trust.

Conclusion and Future Enhancements

Planistry integrates advanced AI methods with intentional, human-centered design to rethink how couples find wedding vendors. The model’s ability to translate visual inspiration into personalized matches is supported by practical filters like location and availability, ensuring recommendations are both relevant and actionable.

Interactive features, including the Planistry Companion chatbot, give users meaningful control over their experience, helping them define their vision and make confident decisions. Transparency, fairness, and privacy have been central considerations throughout the design to foster trust and support a diverse, equitable marketplace.

Looking forward, key opportunities for growth could include:

  • Dynamic availability integration to sync vendor calendars and provide real-time booking options

  • Quote automation for faster, more convenient vendor interactions

  • Broader data inclusion to improve model accuracy and representation, including expanding style and cultural diversity

  • Ongoing user feedback loops to continuously refine recommendations and user experience

This balanced technical and user-focused foundation positions Planistry to evolve as a powerful tool that simplifies wedding planning while respecting users and vendors alike.