Inspiration Palmistry is a timeless conversation starter and a fun way to explore personality, but traditional readings are often inaccessible or expensive. Existing entertainment apps usually rely on random number generators that feel fake and disconnected from reality.
We wanted to bridge the gap between mysticism and technology. Our goal was to create an "AI Fortune Teller" that actually sees your hand using computer vision, providing a personalized, consistent, and socially shareable experience akin to modern personality tests (like MBTI), but powered by the shape of your hand.
What it does PalmSight AI transforms your smartphone camera into a digital palm reader:
Visual Guidance: An AR-style overlay guides users to capture the perfect palm photo.
Multimodal Analysis: The app sends the image to Google Gemini 2.5 Flash. The model analyzes physical features (hand shape, finger length) and major lines (Heart, Head, Life lines).
"Palm MBTI" System: We categorize users into 5 Elemental Personality Types (e.g., "Fire Hand - Enthusiastic Leader") based on visual geometry.
Consistent & Creative: It provides a "Lucky Item" of the day and detailed scores for Career, Love, and Health.
Social Sharing: Generates a shareable summary to compare results with friends.
How we built it
Frontend: Built with vanilla HTML5, CSS3, and JavaScript. We optimized for a lightweight, "Zero-Backend" architecture to ensure fast loading and easy deployment on Netlify.
AI Core: The heart of the app is the Gemini 2.5 Flash API. We utilized its multimodal capabilities to process the image directly without intermediate OCR or object detection models.
Prompt Engineering ("Visual Anchoring"): This was our secret sauce. We implemented a strategy to stabilize the LLM's output. We instructed Gemini to strictly classify hand shapes (Square palm + Short fingers = Earth Type) to ensure consistency, while allowing high temperature (creativity) for the daily fortune text.
Challenges we ran into
Consistency vs. Randomness: Initially, the model gave different personality types for the same hand. We solved this by refining the prompt to prioritize "Physical Geometry" for classification, effectively anchoring the AI's reasoning to visual facts.
Camera & Privacy: Handling camera streams across different mobile browsers (iOS Safari vs. Android Chrome) was tricky. We implemented a robust fallback system.
Lighting: Poor lighting made lines hard to see. We added a "Spotlight" UI overlay to subconsciously guide users to center their hands and improved image pre-processing.
Accomplishments that we're proud of
The "Palm MBTI" Concept: Successfully turning a traditional fortune-telling activity into a modern, categorical personality test.
Single-File Architecture: Creating a fully functional, visually rich AI application contained entirely within a single .html file.
Visual Feedback: The UI feels magical with scanning animations and "breathing" guidelines.
What we learned
Prompting for Vision: We learned that multimodal models need specific instructions on where to look (e.g., "Analyze the curvature of the Life Line relative to the thumb") to get accurate results.
Speed of Gemini 2.5 Flash: Its low latency is crucial for consumer entertainment apps, making the "scan-to-result" flow feel instantaneous.
What's next for PalmSight AI
Multi-turn Chat: Allowing users to ask follow-up questions to the "AI Master" about specific lines.
Couple Match: Scanning two hands to analyze compatibility.
History Tracking: Saving readings to track luck scores over time.
Built With
- computer-vision
- css3
- google-gemini
- html5
- javascript
- prompt-engineering
Log in or sign up for Devpost to join the conversation.