Travel Pose Helper

An AI-powered web app designed to assist travelers in capturing perfect photo poses using images of their current location and help users to compare their image with the example image.

Comment

Inspiration

We noticed many tourists struggle to take well-posed photos at famous landmarks. They often don’t know how to pose or how to capture creative angles, especially when traveling solo. We wanted to build a tool to make posing fun, smart, and personalized.

What it does

Our app allows users to upload or capture a photo at a landmark. Using AI, it generates personalized pose ideas by analyzing the image’s visual context and metadata. Users can:

  • Search reference poses by keywords such as location, gender, objects or by images of the desired background you want to take picture with.
  • Compare reference poses with your photo side-by-side.
  • Receive visual feedback and tips to improve the next shot.

How we built it

Frontend:

  • Built with React, Tailwind CSS, and Vite for rapid UI development.
  • Responsive, mobile-friendly experience with animations and pose overlays.

Backend:

  • Built with Python + FastAPI, deployed via Docker.
  • Integrated MongoDB to store user photos, pose metadata, and filters.
  • Used BLIP to extract tags from image.
  • Used EfficientNet fine-tuned on the Google Landmarks dataset to extract vector from image
  • Used MediaPipe to extract body keypoints
  • Built a rule-based model on top of these keypoints to compute a pose matching score and generate feedback to help users adjust their pose

HP AI Studio was central to our development.

  • We used it to fine-tune pose classification models using a labeled dataset.
  • We tracked multiple experiments, hyperparameters, and model versions with built-in MLflow integration.
  • We exported and deployed the best-performing model as a REST endpoint directly through the platform.

Challenges we ran into

  1. Posing Guidance for Solo Travelers: Many solo travelers struggle to know how to pose naturally or creatively at landmarks, often ending up with awkward or repetitive photos. The lack of a photographer or guide makes it difficult to get real-time feedback or inspiration.

Solution: We addressed this by building an AI system that analyzes the uploaded photo’s context—such as background, subject orientation, and landmark type—and uses a trained model to suggest personalized poses. By combining visual keyword extraction from Vision Model and a custom classifier trained in HP AI Studio, we deliver suggestions that feel relevant and engaging, even without human help.

Accomplishments that we're proud of

  • Delivered an end-to-end AI solution with frontend, backend, and ML capabilities.
  • Successfully used HP AI Studio to train and deploy a working model in under 48 hours.
  • Made posing fun, engaging, and accessible for solo travelers and tourists worldwide.

What we learned

  • HP AI Studio simplifies the entire ML lifecycle, from data ingestion to deployment.
  • Custom model training requires iterative tuning—and MLflow helped us manage that seamlessly.
  • Real-world AI applications benefit from tight integration between UX and ML outputs.

What's next

  • Expand this idea to other domains where correct body posture is essential, such as sports training and yoga practice
  • Incorporate Generative AI to create a visual representation of the user in their desired pose
  • Launch a mobile version with offline pose detection.

Built With

Share this project:

Updates