TECHNOLOGY > AI COACHING ENGINE
HomeTechnologyAI Coaching Engine

How Vora's AI
Actually Works

Most health apps say “AI-powered” and leave it at that. We think you deserve to know what happens between your wearable collecting a data point and Vora telling you what to do today. Here is the full pipeline, including what the AI cannot do.

THE PIPELINE

From Raw Data to Daily Decisions

Every morning, your recommendations pass through five distinct stages. Each stage transforms the data, adds context, and narrows the output until a concrete plan emerges. Nothing is random. Nothing is generic.

Data Ingestion
Normalization
Trend Analysis
Decision Engine
Daily Output
01

Data Ingestion

Raw data flows in from connected wearables (Apple Watch, Oura Ring, Garmin, WHOOP), Apple Health, and manual logs. Each source provides different signals at different frequencies and formats.

Heart rate every 5 seconds from your watch. Sleep stages every 30 seconds from your ring. Steps aggregated hourly from your phone. Workout logs with sets, reps, and RPE from manual entry.

02

Normalization

Unit conversion, timezone alignment, and conflict resolution. Different devices report the same metric in different units, at different times, with different definitions.

HRV as RMSSD vs LnRMSSD. Calories as kcal vs kJ. Sleep onset in local time vs UTC. When two devices disagree on the same metric, the reconciliation engine resolves the conflict before the AI ever sees it.

03

Trend Analysis

Rolling windows at 7, 14, and 30 days detect short-term fluctuations, medium-term patterns, and long-term trajectories. Single-day anomalies are flagged but not overweighted.

A single night of poor sleep triggers a conservative day. Three consecutive poor nights shifts the 7-day trend and adjusts your entire week. A 30-day HRV decline triggers a deload recommendation.

04

Decision Engine

A hybrid system combining rule-based logic for safety constraints with ML models for personalization. Rules enforce hard limits. ML learns your individual response patterns.

Rules: never program heavy deadlifts after a readiness score below 50. ML: learns that YOUR HRV recovers faster after upper-body sessions than lower-body sessions and adjusts split sequencing.

05

Daily Output

The final output: a workout plan calibrated to your readiness, nutrition targets adjusted to your training and recovery, and a recovery recommendation based on accumulated load.

You wake up and open Vora. Your workout is already built. Your calories and macros are set. Your recovery guidance is clear. No manual calculation. No guesswork.

READINESS SCORING

How Your Readiness Score Is Calculated

Your readiness score is a weighted composite on a 0-100 scale. It is not based on a single metric or a single night. Each input contributes a specific percentage, and the weights reflect how strongly each factor predicts next-day performance in peer-reviewed research.

Critically, HRV is evaluated as a trend, not a single reading. One low HRV night after a hard training day is expected. Three consecutive nights of declining HRV is a genuine recovery signal.

HRV Trend (7-day)25%

Not a single reading. The direction and magnitude of your HRV over the past week relative to your personal baseline.

Sleep Quality & Staging22%

Total duration, deep sleep percentage, REM percentage, sleep efficiency, and number of awakenings. Weighted by reconciled data when multiple sources exist.

Resting Heart Rate Deviation18%

How far your overnight resting heart rate deviates from your 14-day rolling average. Elevations of 3+ bpm signal incomplete recovery.

Body Temperature Deviation12%

Deviation from your personal temperature baseline. Elevated readings can indicate illness onset, overtraining, or hormonal shifts.

Previous Day Training Load13%

Volume, intensity, and muscle group demand from yesterday. Heavy lower-body sessions carry more residual fatigue than isolation work.

Subjective Readiness Input10%

Your self-reported energy, motivation, and soreness. The AI trusts your subjective input because you know things your wearable cannot measure.

Total: 100%. Weights are not static. If a data source is missing, its weight is redistributed proportionally across the remaining inputs. The score always reflects the best available information.

ADAPTIVE TRAINING

How Your Workouts Adapt

Your readiness score maps directly to one of four training intensity tiers. This is not a suggestion. It is a calibrated adjustment that protects you on bad days and pushes you on good ones.

READINESS
85 - 100

Full Intensity

Progressive overload is active. Full prescribed volume with intensity progression. This is where gains happen.

Example: If your program calls for 4x6 back squat at 82.5% 1RM, you get exactly that. If you have been hitting all reps consistently, the AI may nudge weight up by 2.5%.

READINESS
70 - 84

Reduced Volume

Intensity is maintained but volume drops. You still lift heavy, just fewer sets. This preserves strength stimulus while reducing total fatigue cost.

Example: 4x6 becomes 3x6 at the same weight. Accessory work is trimmed. The session is 15-20 minutes shorter.

READINESS
60 - 69

Technique Focus

Moderate load with emphasis on movement quality. Good for reinforcing motor patterns without accumulating significant fatigue.

Example: Back squat drops to 3x5 at 70% 1RM with tempo emphasis. Accessory work shifts to mobility and stability drills.

READINESS
Below 60

Active Recovery or Rest

The AI recommends skipping resistance training entirely. Options include light walking, yoga, stretching, or complete rest depending on the specific readiness breakdown.

If sleep was the primary driver, it prioritizes rest. If training load was the driver but sleep was fine, it may suggest light movement to promote blood flow and recovery.

Key principle: The AI never skips a tier. A readiness score of 72 never triggers full-intensity programming, even if yesterday was a rest day. The tiers exist to prevent the most common training mistake: going too hard when your body has not recovered.

ADAPTIVE NUTRITION

How Your Nutrition Adapts

Static calorie targets assume every day is the same. They are not. Your energy expenditure, recovery needs, and hormonal context change daily. Vora adjusts your nutrition targets in response to what actually happened, not what was planned.

Post-Poor-Sleep Night

Protein target increases by 10-15%

Sleep deprivation impairs muscle protein synthesis and increases cortisol. Higher protein intake partially offsets the anabolic resistance that follows a bad night of sleep.

Post-Heavy-Training Day

Caloric surplus of 200-400 kcal

Heavy resistance training elevates energy expenditure for 24-48 hours through EPOC and repair processes. The surplus supports recovery without requiring manual tracking adjustments.

Deload Week

Caloric target drops to maintenance

Training volume is reduced, so the energy demand drops. Maintaining a surplus during a deload leads to unnecessary fat gain without the training stimulus to drive adaptation.

Cycle Phase Awareness

Macro ratios shift by cycle phase

During the luteal phase, basal metabolic rate increases and carbohydrate tolerance shifts. Vora adjusts caloric targets upward slightly and modifies carb/fat ratios when cycle tracking data is available.

Adjustments stack. A heavy training day after a poor night of sleep triggers both a protein increase and a caloric surplus. The AI resolves overlapping triggers and produces one coherent set of daily targets.

PERSONALIZATION

What “Personalized” Actually Means

Every fitness app claims personalization. Most of them mean a questionnaire at signup. Vora means something different: a model that continuously learns your individual physiology and progressively improves over time.

Days 1-14~45% accuracy

Baseline Learning

Vora establishes your personal norms. Recommendations are conservative and rely more on rule-based logic than personalization. The AI is observing, not yet optimizing.

2 Weeks - 1 Month~68% accuracy

Pattern Recognition

The model begins detecting your individual response patterns. It learns how your HRV responds to training, how your sleep affects next-day readiness, and where your recovery bottlenecks are.

1 - 3 Months~82% accuracy

Refined Personalization

Enough data exists to capture weekly rhythms, lifestyle patterns, and training response curves. Recommendations become noticeably more accurate and individually calibrated.

3+ Months~92% accuracy

Deep Adaptation

Seasonal patterns, stress cycles, and long-term periodization effects are factored in. The AI has seen you at your best, worst, and everything in between. It knows your patterns better than you do.

Baseline Learning Period

The first two weeks are observation. The AI collects data without aggressive optimization. It maps your typical HRV range, your normal sleep patterns, your resting heart rate floor, and your subjective energy cycles.

Individual Comparison, Not Population

An HRV of 35ms is concerning for someone whose baseline is 55ms. It is perfectly normal for someone whose baseline is 38ms. After the baseline period, every metric is compared against YOUR history, not age-sex averages.

Progressive Model Refinement

The model never stops learning. Seasonal changes, training phase shifts, life stress periods, and even timezone changes are incorporated into your personal model as they accumulate.

GRACEFUL DEGRADATION

When Data Is Missing

Real life is messy. You forget to charge your watch. Your ring loses Bluetooth connection. You skip logging a workout. A useful AI must handle incomplete data without falling apart or making reckless assumptions.

HRV Data Missing

Primary action: Increase weight on sleep quality and RHR deviation
Fallback: If both HRV and sleep are missing, default to 65 readiness (technique focus tier)

Sleep Data Incomplete

Primary action: Use last-known sleep pattern with 15% confidence penalty
Fallback: Prompt for manual sleep estimate. If no input within 2 hours, assume conservative 6-hour night.

No Wearable Data (device not worn)

Primary action: Rely entirely on subjective input and previous-day training load
Fallback: Default to reduced-volume tier with option to override if you feel fully recovered.

Training Log Not Updated

Primary action: Estimate from heart rate and activity data during the training window
Fallback: If no activity data exists, assume rest day for load calculation. Next workout is not auto-progressed.

New User (no historical data)

Primary action: Use age, sex, training experience questionnaire for initial calibration
Fallback: Conservative programming with moderate volume and intensity. Baseline learning begins immediately.

Core principle: When data is missing, the AI always defaults to conservative recommendations. It will never push you harder because it lacks information. Uncertainty maps to caution, not aggression.

HONEST LIMITATIONS

What the AI Does Not Do

Trust requires honesty about boundaries. Here is what Vora's AI explicitly does not claim to do, and why those boundaries exist.

Does Not Diagnose Medical Conditions

Vora is a coaching tool, not a diagnostic tool. It does not identify diseases, interpret lab results, or provide clinical assessments. If it detects persistent anomalies like sustained elevated resting heart rate, it recommends consulting a healthcare provider.

Does Not Replace Professional Medical Advice

Vora provides fitness and wellness recommendations based on your data. It does not replace the judgment of physicians, dietitians, or licensed health professionals. For medical concerns, always consult a qualified professional.

Does Not Sell or Share Your Data

Your health data is never sold, shared with third parties, or used for advertising. Processing happens on-device or in your private encrypted cloud instance. You can export or delete your data at any time.

Does Not Use Population Stereotypes

After your baseline period, the AI compares you against your own history. It does not assume that a 30-year-old male should have a specific HRV or that a 45-year-old female should sleep a certain amount. Your data defines your norms.

Does Not Guarantee Outcomes

Vora optimizes your training, nutrition, and recovery based on available data. Results depend on consistency, adherence, genetics, and factors outside the scope of any AI system. The AI improves your odds, not your certainty.

Does Not Override Your Judgment

Every recommendation can be overridden. If the AI says rest but you feel ready to train, you can proceed. Vora logs the override and adjusts future recommendations based on the outcome. You are always in control.

Multi-Device Data Reconciliation

The AI coaching engine depends on accurate input data. When you connect multiple wearables, each reports different numbers for the same metric. Vora's data reconciliation layer resolves these conflicts before the AI ever sees the data, ensuring that your readiness score and recommendations are built on a clean, unified timeline rather than conflicting device outputs.

Deep dive: How Vora reconciles multi-device data

What is Vora?

Vora is an AI health coaching app for iOS that integrates data from your wearables, Apple Health, and manual inputs to deliver daily personalized workout programming, nutrition targets, and recovery recommendations. It is built for people who take their health seriously but do not want to spend hours interpreting data and building their own plans.

The AI coaching engine described on this page is the core intelligence layer that powers every recommendation in the app. It processes your data, learns your patterns, adapts to your life, and gives you a clear, actionable plan every morning.

Adaptive Workouts
Programmed daily based on readiness
Dynamic Nutrition
Targets that adjust to your day
Recovery Intelligence
Know when to push and when to rest
Multi-Device Support
Apple Watch, Oura, Garmin, WHOOP

Frequently Asked Questions

How does Vora decide my workout for the day?
Vora calculates a readiness score each morning from six weighted inputs: your HRV trend over the past 7 days, sleep quality and staging from last night, resting heart rate deviation from your 14-day baseline, body temperature deviation, yesterday's training load, and your subjective readiness input. The composite score maps to one of four intensity tiers that determine your training volume, intensity, and exercise selection. A score of 85+ means full progressive overload. Below 60 means active recovery or rest.
How long does Vora take to learn my patterns?
The baseline learning period is approximately two weeks. During this time, Vora establishes your personal norms for sleep, HRV, heart rate, and activity patterns. Recommendations are conservative during baseline. After 30 days, pattern recognition improves significantly as the model detects your individual response curves. By 90 days, seasonal patterns, stress cycles, and long-term training adaptations are factored in. The model continues refining indefinitely.
What happens if my wearable data is incomplete?
Vora degrades gracefully. If HRV data is missing, it redistributes that weight to sleep quality and resting heart rate deviation. If sleep data is incomplete, it uses your most recent pattern with a confidence penalty and may prompt you for a manual estimate. If no wearable data is available at all, the AI relies on subjective input and previous-day training load. It always defaults to conservative recommendations when data is absent.
Does Vora use population averages or my own data?
After the initial two-week baseline, Vora compares you exclusively against your own historical data. An HRV of 45ms might be excellent recovery for someone whose baseline is 40ms, and a red flag for someone whose baseline is 65ms. Population averages are only used during the initial baseline when no personal data exists yet. Once your baseline is established, all comparisons are individual.
Can Vora diagnose health conditions?
No. Vora is a coaching and wellness tool, not a medical device. It does not diagnose conditions, interpret lab results, or provide clinical assessments. If it detects persistent anomalies in your data, such as sustained elevated resting heart rate or a multi-week HRV decline that does not respond to rest, it will recommend consulting a healthcare provider. It will never attempt to explain the cause.
How does the AI handle nutrition for women differently?
When cycle tracking data is available through Apple Health or manual input, Vora adjusts nutrition targets by cycle phase. During the follicular phase, standard targets apply. During the luteal phase, basal metabolic rate typically increases, so caloric targets rise slightly and carbohydrate-to-fat ratios shift. During menstruation, the AI emphasizes iron-rich food suggestions and adjusts training intensity expectations downward. These adjustments are based on published research on female athletic performance across the menstrual cycle.

Explore More Technology

Technology Overview
How Vora processes your data
Data Reconciliation
Multi-device conflict resolution
Sleep Accuracy
Deep dive into sleep staging
Biology & Health Score
Your unified health score

Let the AI do the thinking.

Connect your wearables, let Vora learn your patterns, and wake up every morning to a plan that is built for your body, your recovery, and your goals.

Download FreeSee Plans

Related Reading

Your Devices Disagree. Here's Why That Matters.
Technology deep dive
HRV Explained: What Heart Rate Variability Tells You About Recovery
8 min read
The Science of Sleep and Athletic Performance
9 min read
Vora Technology: How We Process Your Health Data
Overview

Continue Reading

Data ReconciliationSleep AccuracyHeart Health MetricsNutrition AITraining Load
Back to Technology