2026 / Mist / Indie App
Designing input for an AI powered soundscape engine, from feeling to frequency
/ overview
What if relaxation technology could adapt to how you feel?
Existing sound wellness products rely on fixed libraries of pre-recorded soundscapes.
What if we could personalize relaxation by adapting sound to a person's emotional state?
/ solution
I built Mist, an AI powered and emotion aware sound engine
try it here: www.mistsound.com
user input
the user describes how they are feeling to mist
LLM extraction
mist uses AI to map those emotions into researched psychoacoustic parameters
soundscape render
the sound engine then uses those parameters to render a personalized soundscape
/ technical architecture
AI? How does the product backend work?
Mist does not generate audio. It generates recipes from a pre-curated sound library.
/ the problem
What v1 got wrong :(
thin user inputs = poor personalization
/ whoops. how did I discover this?
the insight surfaced when testing with the first 8 users:
the 8 soundscape recipes showed 43% similarity in recipe,
and utilized only 13% of all audio components in the sound library!
/ digging deeper
why was this happening? i had to investigate the AI pipeline
intake flow in v1
the initial intake flow asked users to articulate their emotional state in their own words, with little guidance.
after intake:
how does the LLM + system prompt break this input down?
but what if:
the user input does not have signals of a particular parameter?
/ testing insights
the data that user testing revealed:
undetected parameters repeatedly defaulted to preset values
key insights:
/ initial assumption
users' descriptions of their state of mind would naturally contain the emotional signals the AI needed
/ reality discovered
users' unstructured + natural input often lacked the signals needed to score all four parameters
/ rephrasing the problem
how might we
bridge the gap between what users say and what the system needs?
/ iterating
idea one:
effort reducers to help the user spend less time thinking
emotion cues
select how you feel
sentence prompts
complete a thought
imagery prompts
choose an atmosphere
common situations
choose a frame of mind
tradeoffs in this approach:
recognition (that’s me) is a lower bar for entry than generation (this is what I am thinking)
doesn’t rely on raw insights from the user, which is the MVP and differentiator
/ iterating
idea two:
breaking the intake flow into multiple questions targeting specific parameters
intake start
open ended prompt targeting valence, the broadest schema.
fixed follow ups
pre-written follow up questions that target each schema
tradeoffs in this approach:
detailed information about each schema = more accurate guess
highly inconvenient for the user to answer multiple questions
/ iterating
idea three:
generative follow up questions
conditional follow ups
targeted follow ups with generated questions if the system cannot detect few schemas
tradeoffs in this approach:
generative copywriting can be personalized to situation
every schema does not benefit from descriptive input
/ final flow
combining best practices:
a generative flow that combines open-ended & close-ended follow ups
start of the intake
a broad, open-ended prompt to encourage free expression
prompt chips to guide users when they're unsure
voice-first input, as users tend to express more when speaking
conditional follow up 1
open ended follow up asked only if both valence and arousal are absent
why?
both are essential signals
open-ended input is the best way to surface them
designed for edge cases: gibberish or unrelated input
generative copy, based on earlier input
why?
intent & warmth are preferences, not descriptive signals
open-ended input adds unnecessary effort once emotional state is known
conditional follow up 2
close ended follow up asked if either intent or warmth are absent in the input
/ impact
repeating the same prompts with 8 test users
resulted in significantly more diverse and personalized soundscapes.
across 8 users
before
after
recipe similarity
43 %
21 %
library utilization
13 %
56 %