An intelligent fitness-coaching platform designed as more than a pose tracker. The system combines a multi-agent LLM architecture with real-time computer vision to build personalized plans, monitor execution, evaluate progress, and continuously adapt recommendations as the user changes.
SYSTEM ARCHITECTURE9 specialized AI agents + vision feedback loop
INPUTUser profile & progressGoals · body changes · history · performance
→
ORCHESTRATION9-Agent LLM SystemSpecialists coordinate around the user
→
OUTPUTAdaptive Coaching PlanContinuously revised recommendations
01 · MULTI-AGENT INTELLIGENCE
Specialized agents, one coordinated plan.
The LLM layer is structured around nine specialized agents rather than a single generic assistant. Individual agents own focused coaching responsibilities — including areas such as nutrition and workout planning — while the wider system combines their outputs into a coherent user plan.
The plan is not static: recommendations can be revised as the user progresses, performance changes, or new body measurements and conditions are introduced.
02 · REAL-TIME COMPUTER VISION
Form evaluation while the exercise is happening.
The vision component tracks body pose during exercise, evaluates movement and joint alignment, and identifies form issues in real time. Instead of only counting repetitions, the system is designed to produce specific corrective feedback when posture deviates from the expected movement.
That means feedback can respond to details such as an elbow drifting inward or outward, helping the user correct technique during the set rather than after it.
03 · CLOSED FEEDBACK LOOP
Observe → evaluate → adapt.
Training behavior and user progress feed back into the coaching layer. The concept links what the user is prescribed with what the user actually performs, allowing the system to support a more adaptive coaching experience over time.
The project required integrating AI model behavior, computer-vision logic, user-facing feedback, and team-level project execution into one end-to-end product concept.