Custom multimodal datasets
Records, images, audio and device signals, linked across a participant’s timeline and prepared to your specification.
LONGITUDINAL HEALTH DATA FOR AI
Health data over time.
Built for your model.
Clinical records, device signals and everyday experience from the same people over time. Custom collections for AI training and evaluation.
In development. Scoping initial pilot studies.
WHAT YOU CAN COMMISSION
From real health histories
to training and evaluation.
Commission the collection, preparation or evaluation your model needs. Define the sources, review and delivery together.
Longitudinal collections
Repeated measurements and follow-ups, scoped around the people, signals and duration your research needs.
Expert-reviewed training examples
Source-grounded demonstrations, annotations and reference answers, with qualified review defined for the task.
Custom evaluations
Held-out cases, task-specific rubrics and scoring criteria to assess the capabilities your team wants to improve.
THE COLLECTION EXPERIENCE
Better data starts
with people coming back.
The LifeFrom app brings records, measurements and everyday context into a study-defined schedule.
Upload an export, share a record or answer a check-in. Guided capture and follow-ups make room for what happens between visits.
Explore the consumer experienceTalk. Capture. Connect.Follow up.
LIFEFROM DEVICES / STARTING WITH THE BAND
Built to capture life.
Designed for learning.
The first device in our planned research lineup. Connected by LifeFrom’s proprietary collection software, with reinforcement learning and evaluation in mind from the outset.
DEVICE 01
LifeFrom Band
Wearable signals. A voice in the moment.
SIGNALS & CAPTURE
- 01 Heart rate
- Average, minimum, maximum & resting heart rate
- 02 HRV & oxygen
- Device-derived HRV & blood oxygen estimates / SpO₂
- 03 Activity
- Steps, distance, calorie estimates & activity-intensity minutes
- 04 Sleep
- Timing, duration & estimated deep, light, REM and awake periods
- 05 Temperature & stress
- Temperature readings & device-derived stress scores
- 06 Voice & context
- On-band microphone, recorded audio, timestamps & transcripts
Time-stamped device readings, recorded audio and transcripts, connected in the participant’s timeline.
VOICE-ASSISTED DATA LABELING
The band records the signal.
The person supplies the context.
Meals, exercise, symptoms, medication, supplements and daily routines, described in the participant’s own words. Our study workflow is designed to turn those notes into time-aligned labels, with follow-up questions and participant corrections.
01 / SPEAK · BAND MICROPHONE · 08:32
“Just finished a 20-minute bike ride. Started at 8:10. Felt hard.”
02 / CLARIFY · IN THE APP
Was that indoors or outdoors?
“Indoors, on a stationary bike.”
A button press starts a note. Study check-ins can invite participants to describe an event, then clarify missing details.
03 / ALIGN & LABEL
Reported event window08:10–08:30
Link available readings before, during and after the event.
Heart rate · HRV · activity- Activity
- Stationary cycling
- Duration
- 20 minutes
- Effort
- Hard, self-reported
AI-extracted · awaiting confirmation
Raw source data. Structured labels.
Original audio and source device records, delivered alongside transcripts, time-aligned tags and annotations.
Our collection stack.
Direct band-to-app integration. LifeFrom links device readings, audio and transcripts in its own system.
RL, by design.
Planned task packages pair signal windows, questions, participant corrections and reviewed labels with scoring rubrics for training and evaluation.
Permissions at the source.
Study-specific consent and licensing define which records can be used, by whom, and for what training purpose.
Choose the inputs.
Keep the context.
Combine clinical history, measured signals and participant logs. Each source adds a different view of the same timeline.
01
Medical records
History, with its source intact.
Bloodwork, DNA, molecular tests and clinical imaging, scoped with source files, reports and follow-ups.
Explore this data02
Connected measurements
The signals between visits.
Heart rate, HRV, sleep, activity, temperature and weight. Device signals with timestamps, source metadata and coverage.
Explore this data03
Life, in context
Ask the question the record cannot.
Audio, photos and video. Food, medication, supplements and symptoms, logged through guided conversations and check-ins.
Explore this dataCollection scope depends on consent, device access and study feasibility. Raw signals where the source supports them.
Explore all data typesThe observations.
And the context to use them.
Follow a source-linked record from collection to delivery, with timing, missingness and review decisions intact.
EXAMPLE FOLLOW-UP SCHEDULEONE PARTICIPANT / OVER TIME
DAY 00
Establish a baseline
Record + context
DAY 14
Capture the change
Measurements + check-in
DAY 90
Measure the follow-up
Repeat measurement + follow-up
Time-aligned inputsSource provenanceStudy-defined labels
Scope the collection
Eligibility, geography, collection windows and follow-up schedule, agreed before recruitment.
Prepare the dataset
Source checks, normalized units and timestamps, documented missingness. Annotation and qualified clinical review scoped to the task.
Define the delivery
Versioned JSONL or Parquet, linked media and a dataset card. Training and evaluation splits separated by participant, with task-specific leakage checks.
THE STUDY COMES FIRST
Clear permissions.
Inspectable quality.
Know where each observation comes from, how it was reviewed and which uses are permitted.
What does your model need to learn?
Start with the task. Define the collection together.
Scope a pilot