Architecture

The technology
behind the engine.

Three layers. One closed loop. Every decision VirtueFit makes is grounded in computer vision, constrained optimisation, and a continuously updated knowledge base. No templates. No shortcuts. Just engineering.

01 Vision 02 Intelligence 03 Knowledge Repeat
01 Computer Vision

See the body. Model the body.

A single RGB video stream from your phone camera is transformed into a full 3D reconstruction of your physique. No wearables, no depth sensors, no external hardware.

vision_pipeline.run()
Capture
8–12s RGB video at 30fps
Pose
Landmark detection (33 keypoints)
Depth
Monocular depth inference
Mesh
SMPL parametric body model
Output
3D model, BF%, 14 measurements
Execution: on-device · Latency: <60s · No cloud upload required
FrameworkMediaPipe + custom regression head
Body modelSMPL parametric mesh (6,890 vertices)
Measurements14 circumferences, 3 composition metrics, 2 postural scores
Delta trackingScan-over-scan vertex displacement for visual change detection
02 Inference Engine

Reason. Generate. Adapt.

Scan data, session logs, and user signals feed a constraint-satisfaction engine that generates and continuously re-generates your training and nutrition programme. Every output is a function of your current state, not a static prescription.

engine.inference_cycle()
Inputs
Scan data Session logs RPE signals Meal logs Check-ins
Engine
Constraint
solver
312 variables Multi-objective Realtime
Outputs
Programme Nutrition plan Load targets Meal suggestions Coaching context
Trigger: post-session or schedule conflict · Cycle: <24h · Mode: generative, not rule-based
OptimisationVolume, intensity, frequency, recovery balanced per mesocycle
NutritionBMR from scan → TDEE from activity → surplus/deficit from goal → daily macro targets
PersonalisationSplits, exercise selection, rep ranges, and timing adapt to preference and response
CoachingNatural language interface with full read access to profile, programme, and log state
03 Knowledge Base

Built on evidence. Updated continuously.

The engine does not hallucinate training advice. Every recommendation traces to a foundation of peer-reviewed exercise science, clinical nutrition guidelines, and validated coaching methodology. The knowledge base is versioned, tested, and continuously expanded.

knowledge.source_registry()

Research literature

Peer-reviewed studies in exercise physiology, biomechanics, and sports nutrition

Clinical guidelines

ACSM, ISSN, and NSCA position stands on training, nutrition, and body composition

Training data

Real-world session logs, adherence patterns, and response curves from active users

Continuous updates

Versioned knowledge releases integrated into the engine without downtime

Sources: peer-reviewed · Methodology: evidence-graded · Updates: rolling
GroundingEvery prescription maps to a cited principle or validated heuristic
ValidationAutomated regression tests on output quality after each knowledge update
ScopeExercise science, human physiology, clinical nutrition, coaching methodology

Three layers. One loop. Zero guesswork.

Vision reads your body. Intelligence builds your plan. Knowledge keeps it honest. Every session closes the loop.

Try VirtueFit free