Poseidon

POSEIDON
Poses library
Coaching View
Result Page
Poseidon team
Thank you

Inspiration

Confidence shouldn’t be a luxury, and a camera lens shouldn’t feel like a barrier. Whether it’s a professional milestone or a candid memory, everyone deserves to feel seen. That’s why we created Poseidon, an AI-powered personal directory that transforms your smartphone into a modelling workshop. We aren’t just building a camera app - we’re building the bridge between how you feel and how you look, ensuring that your digital story does justice to the real you. 🎥✨

We believe that photos are the modern way we tell our stories. Some of our most cherished moments are captured in a single frame, yet we noticed a recurring problem: so many people stay behind the lens because they’re afraid of how they’ll look in front of it. You probably know someone who:

📸 The "Designated Photographer": The friend who always volunteers to hold the camera because it feels too scary to join a group photo
💼 The “LinkedIn Ghost”: The talented classmate who avoids networking platforms because they don't have the confidence to take photos for their LinkedIn.
🌴 The "Scenery-Only" Traveler: The person who misses out on capturing memories on holiday because striking a pose in public brings up too much embarrassment.

Why should a lack of “modelling skills” keep you from capturing a lifetime of memories? If photos are about self-expression, why do so many people feel silenced by the shutter? We realised that traditional camera apps focus on filters and editing after the fact, but they do nothing to help you in the moment when the pressure is on.

Our mission is simple. To make sure everyone feels good enough to be in the photo. Poseidon ensures that nobody has to stay behind the lens because everyone deserves to be part of the story 💙

What it does

Poseidon actively teaches you how to pose in front of a camera by combinimg real-time pose detection with LLM pose analysis for personalised guidance.

Key features:

Curated Pose Library ‒ Browse a curated pose library and select a target pose to learn and practice
Real-Time Visual Guidance ‒ A target skeleton will overlay on your screen, using visual prompts like circle landmarks and lines that turn red to green to help align your body
AI Verbal Coaching ‒ Every few seconds, our integrated LLM provides coaching through audio or text-based feedback, like “tilt your chin up” or “fix your posture” to guide you into the right pose
Smart Feedback — Once you're happy with your pose, you can view a 3D comparison between your finished pose and the target.

How we built it

We built Poseidon using NextJS, MediaPipe with Pose Landmarker, and Gemini 2.5 Flash.

NextJS: For the integrated full-stack development experience.
MediaPipe: Open source suite of tools for AI/ML.
Pose Landmarker: A MediaPipe library for live pose detection.
Gemini 2.5 Flash: To generate personalised instructions.

Challenges we ran into

Structuring work: When you only have a weekend to put together a full application, it's much harder to design a clean and extensible architecture which ultimately feeds back into the development experience.
LLM Delay: Originally, our first draft prompt lead to a 60-second delay while the LLM processed our pose data. We had to spend a significant amount of time and thought to reduce prompt size and increase signal in pose data.

Accomplishments that we're proud of

LLM Delay: We were able to reduce the 60-second delay to around 5-seconds, making the live recommendation feature tenable and worthwhile using!
3D Pose Visualisation: We think this is a great special sauce on top of everything else we built, giving the user an entirely new perspective on their pose and allowing them to

What we learned

Maybe FIT2099 was right. At least a little bit of diagramming goes a long way. In the future we'd much more readily work on designing components and their APIs before starting work on implementation.

-Having a problem to solve only gets you five percent of the way. …

What's next for Poseidon

This is just the first frame in our camera roll! 🎞️ Next, we plan to expand our capabilities to handle more specific photos like facial expressions. To keep our database fresh, trendy, and relevant to modern styles, we also want to consider TikTok/Instagram models who pose.

Additionally, we aim to fully build out our frontend by creating a complete camera app looking mockup with our extra features.

Our ultimate goal goes beyond the code. We'll be happy if even one person musters up the confidence to take a photo on holiday, or with their family, that they otherwise wouldn't have. Because at the end of the day, we think memories matter, and you don't have to be camera shy! 🎉📸