SUMEET GOKHALE
Meditation that adapts to how you feel
A mudra recommendation engine that replaces a content library with one right-now practice.
Overview
Part I of Nirvana solved returning. Nudges, streaks and proof loops kept people coming back past the first week. It surfaced a second problem I could not fix with reinforcement alone: people opened the app, faced a library, and closed it again.
This extension replaces browsing with matching. Instead of a catalogue of sessions, Nirvana asks two small questions, how you feel and how long you have, then returns exactly one practice: a mudra, a duration, and a sentence explaining why it was chosen. The product is built around ten traditional hand gestures, each with a documented intent, rendered as schematic vector diagrams so the posture is legible without photography or video.
Problem
Meditation apps front-load a decision at exactly the moment a user has the least capacity to make one. Someone opening the app anxious at 11pm is asked to browse categories, compare durations, and read descriptions. That is cognitive work, and it competes directly with the state they are trying to leave.
Choice is presented as generosity, but for a stressed user a large library is friction. The drop-off does not happen during the session; it happens on the screen before it.
The library model
- User must self-diagnose, then translate that into a search
- Every option looks equally valid, so none feels correct
- Decision cost is highest when the user is least able to pay it
- Success depends on the user already knowing what they need
The matched model
- Two taps, a feeling and a duration, and the system decides
- One recommendation, with the reasoning stated in plain language
- Time of day is inferred, not asked
- The app carries the expertise so the user does not have to
The recommendation engine
The logic is deliberately small and explainable. A feeling maps to a mudra; the time of day acts as a corrective layer. If the clock says it is late and the mapped mudra is an energising one, the system quietly swaps it for a wind-down gesture and tells the user why. Nothing is hidden behind a black box, so every recommendation carries its own justification.
The mudra system
Ten mudras cover the emotional range the onboarding asks about. Each one is defined by the fingertips that meet the thumb, which meant the entire library could be drawn programmatically as a schematic hand rather than shot as photography, which keeps it consistent, accessible and instantly re-skinnable.
| Mudra | Element | Addresses | Hand position |
|---|---|---|---|
| Gyan | Air | Focus, calm, clarity | Index fingertip to thumb, other fingers open |
| Prana | Earth + Water | Energy, vitality | Ring and little fingertips to thumb |
| Vayu | Air | Anxiety, restlessness | Index folded to palm, pressed by thumb |
| Dhyana | Stillness | Deep calm, sleep | Both hands in lap, palms up, thumb-tips touching |
| Shuni | Ether | Patience, anger | Middle fingertip to thumb |
| Prithvi | Earth | Grounding, confidence | Ring fingertip to thumb |
| Buddhi | Water | Clarity, communication | Little fingertip to thumb |
| Anjali | Heart | Centering, gratitude | Palms pressed together at the chest |
| Surya | Fire | Warmth, morning energy | Ring finger folded down, pressed by thumb |
| Apana | Earth | Release, emotional reset | Middle and ring fingertips to thumb |
Design decisions
Ask for a feeling, not a category
Onboarding collects plain-language states like overthinking, burnout and poor sleep, because that is the vocabulary users already have. The translation into practice is the product’s job, not the user’s.
Always show the reasoning
Every recommendation is accompanied by a sentence naming the inputs that produced it. Stated reasoning is what turns an algorithmic guess into something a user is willing to trust and follow.
Draw the gesture, don’t film it
Schematic vector hands render at any size, need no localisation, and highlight only the contact points that matter. Clearer than photography, at a fraction of the production cost.
End-to-end flow
First run takes a user from launch to a practice in eight screens, two of which are skippable. Every return visit starts on Home, where the recommendation has already been calculated.
Interactive prototype
The full flow is a working browser prototype covering onboarding, feeling selection, the recommendation engine, the guided player with its breathing rhythm, and the profile with streaks and achievements. Tap through it below. The session timer runs at four times speed so a fifteen-minute practice can be completed in the demo.
Output
The extension reframes what the app is for. Part I proved that reinforcement brings people back; Part II removes the obstacle that met them when they arrived. Together they form a single argument: retention is a design problem at both ends of the session, not just after it.
- Decision cost cut to two taps. A feeling and a duration replace browsing a library, with time of day inferred rather than asked.
- Ten-mudra library, zero photography. Every gesture is a parametric vector diagram, so the content set scales without a production pipeline.
- Transparent recommendations. Each suggestion states the inputs behind it, including when a late-hour rule overrides the obvious match.
- Working prototype, not a static comp. The logic is real and testable, which made the assumptions falsifiable before any engineering investment.