Conversational Poetry Explorer

TOPLINE
Client
San Antonio Review is an independent literary publisher with a growing catalog of poetry, essays, and visual work. Their current platform allows readers to browse content, but it doesn’t support discovery through emotional or thematic resonance—one of the core ways people naturally engage with poetry.
Project Goal
Design a Poetry Explorer that lets readers describe how they’ re feeling—grief, restlessness, hope, tenderness—and receive poem recommendations that match those emotional states.
The experience should feel conversational, intuitive, and human-centered, not like using a traditional search tool.
My Role
I led the end-to-end design and strategy, from defining the emotional discovery model and information architecture to shaping the conversational interaction patterns and integrating the underlying AI components.
Outcome
I created a conversational Poetry Explorer powered by a Retrieval-Augmented Generation (RAG) approach that blends semantic understanding with thoughtful editorial tone. The system:
Helps readers navigate a 1,000+ poem catalog through emotion and theme
Connects users to work that feels personally resonant, not just categorically relevant
Lightens editorial load by enabling self-guided, emotionally intelligent discovery
Establishes a scalable foundation for future audience engagement tools
PROBLEM
Traditional discovery methods fall short because:

1. Reader Reality
People often search for poetry based on how they feel, not by formal categories.
Emotions like grief, uncertainty, nostalgia, or yearning aren’t reflected in traditional metadata.
2. Catalog Complexity
Over 1,000 poems spanning diverse themes, styles, and tones.
Browsing is linear; emotional pathways are hidden.
3. Editorial Burden
SAR’s editors shoulder most of the work of surfacing relevant pieces.
No scalable way for readers to self-discover meaningful content.
4. Opportunity
Create a discovery experience that responds to human emotional language,
not just keywords or filters.
INSIGHTS
An emotional map, not another method of filtering content

1. Emotional Discovery Is Natural
Readers talk about poetry in emotional terms (“I need something gentle,”
“I’m overwhelmed,” “I want hope that isn’t sentimental”).
They want guidance that feels like a conversation, not a query form.
2. Metadata ≠ Emotional Resonance
“Nature,” “love,” and “loss” are too broad to be useful.
Poems with similar tags often evoke different emotional textures.
3. Editorial Tone Matters
SAR has a distinct voice: clear-eyed, reflective, grounded.
Readers trust recommendations that feel human, not algorithmic.
4. Patterns in Discovery Behavior
Readers rely on editors, social cues, or accident to find new work.
There is no self-directed path based on how they feel right now.
ARCHITECTURE OVERVIEW
1. User Input (Natural Language)
Reader describes their emotional state
(e.g., “I’m grieving but trying to stay hopeful.”)
2. Semantic Understanding
System interprets emotional language
Uses a semantic model to understand tone, themes, and intent
3. Retrieval of Relevant Poems
Searches the poem catalog by meaning, not keywords
Pulls 1–5 poems that align emotionally and thematically
4. Generative Recommendation Layer
A conversational model explains why each poem fits
Uses SAR’s warm, clear editorial voice
Supports multi-turn refinement (“a bit calmer,” “less sentimental,” etc.)

More technical explanation:
User feeling → Semantic representation → Meaning-based retrieval → Editorial-style recommendation
4 steps:
Interpretation
Converts emotional language into a semantic understanding of mood & intent
Semantic Search
Finds poems with similar emotional + thematic signals
Contextual Reasoning
System evaluates which poems resonate and why
Conversational Response
Presents recommendations in SAR's literary voice
Supports multi-turn refinement and deeper exploration
EXPERIENCE BEFORE AND AFTER
Before
Readers browse by category, chronology, or author
Emotional nuance is not searchable
Discovery depends heavily on editorial curation or chance
Navigating 1,000+ poems feels overwhelming and unstructured

After
Readers describe how they feel (“tender but uncertain”, “quietly hopeful”)
System interprets emotional signals using semantic understanding
Retrieves poems with similar emotional and thematic textures
Provides warm, editorial-style explanations
Supports multi-turn refinement (“more contemplative”, “less sentimental”)
DATA MODELING & TAXONOMY
1. Why a Taxonomy Was Needed
SAR’s poems are rich and varied, but their emotional layers are not captured in typical tags
A structured model enables consistent semantic interpretation
2. What the Taxonomy Includes
Themes: conceptual topics (e.g., “Legacy & Impact”, “Nature & Emotion”)
Moods: emotional tones (e.g., “Tender”, “Somber”, “Hopeful”, “Yearning”)
Supporting signals: summaries, emotional keywords, contextual notes
3. How It Informs the System
Serves as a conceptual foundation, not strict rules
Helps the LLM understand emotional patterns
Enhances semantic retrieval with richer context
CONVERSATION FLOW
1. User expresses a feeling
“I’m grieving but don’t want anything sentimental.”
2. System interprets emotional intent
Identifies the emotional texture behind the words
Understands tone (“unsentimental grief”) and desired direction
3. Retrieves meaningful candidates
Uses semantic similarity to find poems with matching emotional signals
4. Composes a warm, editorial recommendation
Highlights why specific poems resonate
No clinical language; no algorithmic explanations
5. Invites refinement
“Would you like something gentler? Darker? More hopeful?”
Supports multi-turn discovery in a natural, conversational way
IMPACT & FUTURE OPPORTUNITIES
Impact on Readers
Opens new pathways to find poems that feel personally meaningful
Turns emotion into a navigable discovery dimension
Makes the literary catalog more accessible, inclusive, and intimate
Impact on SAR Editorial & Operations
Reduces manual curation load
Provides a scalable way to surface back-catalog work
Enhances audience engagement without increasing staff time
Impact on the Platform
Establishes a foundation for AI-driven discovery features
Creates a new interaction model (conversational, emotional, semantic)
Future Opportunities
Integration into SAR’s website as a live reader tool
Extension to essays, interviews, and multimedia work
Voice-based or mobile poetry exploration
Deeper personalization based on reading history
Expansion into a broader “literary discovery” assistant
↑Back to Top