Shangai
ShangAI is a revolutionary, AI-powered browser platform designed for home fitness enthusiasts, martial arts students, and remote training partners who want to elevate their physical performance without investing in costly specialized hardware. Traditional workout applications and pre-recorded exercise videos fail to deliver an interactive experience: they provide zero real-time posture analysis, leaving users unable to detect form errors, which ultimately leads to muscle compensation, reduced training efficiency, and a heightened risk of injury. Furthermore, solo home workouts suffer from high user drop-off rates due to social isolation and a lack of ongoing motivation. ShangAI solves these challenges by transforming any standard webcam into an intelligent motion-tracking coach. Powered by MediaPipe 3D pose detection with sub-50ms latency, the system precisely analyzes joint alignment, movement speed, and execution accuracy across Taekwondo strikes, boxing combinations, and Yoga/meditation poses—delivering instant visual feedback. To ensure complete privacy, all computer vision processing is executed locally on-device. Beyond solo practice, ShangAI features a real-time online P2P multiplayer engine. Users can challenge friends around the world to live head-to-head duels, test precision against each other, and dominate real-time global leaderboards. With planned expansions into additional sports, combat arts, and physical training disciplines, ShangAI turns any web browser into an engaging, interactive, and highly competitive personal training arena.
Preliminary score
- Value for the user17.5 / 25
A specific problem and a clear outcome from solving it.
- Prototype and use of AI21.0 / 30
A working core scenario, and what AI actually contributes to the result.
- Quality testing12.0 / 20
Tests, examples of failures, and comparison with the current approach.
- Feasibility10.0 / 15
Data requirements, running costs and a clear next step.
- Originality6.0 / 10
A meaningful difference from common solutions.
What the judges wrote
- GPT-6 AstraPreliminary72/100
- Strengths
- Your design makes pose-estimation AI central to gameplay while keeping camera frames on-device. Your README provides concrete solo, ghost, and live-match workflows, with unusually clear deployment, privacy, and network limitations. Your reported 120-repetition test, practitioner labeling, and specific framing/backlighting failures provide useful early validation.
- To improve
- Your accuracy and latency claims would be stronger with labeled examples, per-technique error counts, measurement methods, and smoke/clash test results. Your club pilot should include unseen participants and compare against coach-led or video-only practice, measuring improvement and false acceptance of incorrect form. Your deployment plan should quantify signaling and potential TURN-relay costs and test live matches across restrictive networks.
- Claude Opus 5.5Preliminary61/100
- Strengths
- Your README describes a real, deployed architecture rather than a mock-up. On-device MediaPipe pose tracking drives solo challenges, ghost matchups and live head-to-head rounds, and WebRTC data channels send only landmarks between peers. Your privacy design is a genuine advantage: no frames leave the browser, and the README is honest that no external AI service is called. You also report concrete metrics with stated conditions: 120 reps, 88% agreement with a practitioner, 95% rep counting, latency figures, and where tracking degrades (out of frame, backlighting, depth limits of a 2D webcam).
- To improve
- Your testing claims need supporting evidence: publish the labelled test set or a per-technique confusion table, and show specific failed reps and why they failed. The write-up also says 4 techniques but lists only 3. Compare your feedback against how users train today, for example video-only practice versus a coach, so the outcome is measurable rather than asserted. State running costs explicitly (Vercel, Upstash, and TURN relay costs if you add one), and replace marketing language like 'revolutionary' with evidence of user benefit, such as early results from the planned Baku Taekwondo club pilot.
The team
Najaf Guluyev
data
Amrah Mikayilov
developer
İbrahim HüseynovLead
product