FAC-MOCAREV-001 · IWR6843AOP · Phase I Pilot

Simulation
Lab.

Generate 24-hour behavioral datasets for the 12-room Mount Olivet Careview Home pilot. Launch simulations, monitor the AWS pipeline end-to-end, and verify data arrives on resident dashboards.

36

devices

1 fps

per device

0.09M

frames / device / day

0.1 GB

parquet / day / pilot

6

clinical scenarios

Ella AI Nurse Assistant

www.ellamemory.com

00

Simulation Console

Select a scenario · configure parameters · launch · watch the pipeline

Normal Day

BASELINE

18 fps baseline — typical memory-care resident patterns throughout the day.

● SELECTED

Fall Event

CLINICAL

Bathroom fall at 03:00 — AmbientPosture "On Floor", prolonged floor occupancy.

Night Wandering

CLINICAL

High entry + living room activity 01:00–04:00. Agitation or elopement-risk pattern.

Isolation Day

CLINICAL

Near-zero activity all zones all day. Early withdrawal or illness — triggers alert.

Sundowning

CLINICAL

Extreme living room + entry spike 16:00–19:00. Classic late-day agitation pattern.

Care Visit Day

WORKFLOW

Structured check-ins at 10:00, 14:00, 18:00 — entry + living room engagement spikes.

Date

Rooms

Pipeline Status

Generator

Python / Lambda

idle

S3 raw/

Parquet frames

idle

Golden λ

UNLOAD combine

idle

Athena

View refresh

idle

ellamemory

Dashboard live

idle

01

Pilot Roster

12 rooms · FAC-MOCAREV-001 · Mount Olivet Careview Home

301

MOCAREV-0001

ella →

A

B

C

raw —

golden —

302

MOCAREV-0002

ella →

A

B

C

raw —

golden —

303

MOCAREV-0003

ella →

A

B

C

raw —

golden —

304

MOCAREV-0004

ella →

A

B

C

raw —

golden —

305

MOCAREV-0005

ella →

A

B

C

raw —

golden —

306

MOCAREV-0006

ella →

A

B

C

raw —

golden —

307

MOCAREV-0007

ella →

A

B

C

raw —

golden —

308

MOCAREV-0008

ella →

A

B

C

raw —

golden —

309

MOCAREV-0009

ella →

A

B

C

raw —

golden —

310

MOCAREV-0010

ella →

A

B

C

raw —

golden —

311

MOCAREV-0011

ella →

A

B

C

raw —

golden —

312

MOCAREV-0012

ella →

A

B

C

raw —

golden —

02

Activity Preview

Normal Day scenario · 24-hour behavioral model

Compare with

Living Room

Bathroom

Entry

opacity = activity intensity · each column = 1 hour

Night 22:00–06:00

Asleep, low activity
3× overnight bathroom trips (01h, 03:30h, 05h)
Entry near-zero; no staff visits

Day 06:00–15:00

Morning care: hygiene + standing
Absent at dining 07:45–08:15, 11:45–12:30
Staff check-ins every ~2h on entry zone

Sundowning 15:00–19:00

Living room activity spikes 25–40 counts
Entry zone: repeated door-checking bursts
Bathroom mostly empty during agitation

03

Device Configuration

3 zones · 1 device per zone · IWR6843AOP mmWave radar

Living Room

Device IDDEV-MOH-301-A

Mountwall_mount_6m

Zone keyliving_room

FPS1

SuffixA

Zone color0,113,227

Bathroom

Device IDDEV-MOH-301-B

Mountroof_mount

Zone keybathroom

FPS1

SuffixB

Zone color180,83,9

Entry

Device IDDEV-MOH-301-C

Mountwall_mount_corner

Zone keyentry

FPS1

SuffixC

Zone color91,33,182

04

How to Run

scripts/generate_sim_frames.py · ambientcloud repo · Python 3.12+

STEP 01Authenticate to AWS

eval "$(./scripts/resolve_env.sh dev)"
export AWS_PROFILE=pilot AWS_DEFAULT_REGION=us-east-1

# Verify access
aws sts get-caller-identity

STEP 02Dry-run (validates schema, prints frame counts, skips S3 upload)

python3 scripts/generate_sim_frames.py --fps 18 --dry-run
# Output: 36 devices × 1,555,200 frames × 160 bytes ≈ 8.5 GB raw
#         Parquet+ZSTD: ~197 MB for full pilot · day

STEP 03Generate full 12-room pilot at 18 fps

python3 scripts/generate_sim_frames.py --fps 18
# Writes to s3://ambient-dev-parquet-<acct>/raw/date={today}/...
# One file per device: raw/date=.../facility=.../subject=.../device=.../sim_frames.parquet

STEP 04Generate with a clinical scenario

python3 scripts/generate_sim_frames.py --fps 18 --scenario fall
# Available: normal | fall | wandering | isolation | sundowning | care_visit
# Modifies AmbientPosture, AmbientActivityCounts per scenario model

STEP 05Trigger golden combine (merges 3 zone files per room)

aws lambda invoke \
  --function-name ambient-dev-golden-combine \
  --payload '{"date":"2026-05-19"}' \
  --cli-binary-format raw-in-base64-out \
  /tmp/golden-out.json

cat /tmp/golden-out.json
# {"date":"2026-05-19","combined":12,"skipped":0}

STEP 06Verify golden files + Athena smoke query

aws s3 ls s3://ambient-dev-parquet-<acct>/golden/date=2026-05-19/ --recursive

aws athena start-query-execution \
  --query-string "SELECT COUNT(*) FROM ambient_dev_raw.golden_frames WHERE date='2026-05-19' AND facility='FAC-MOCAREV-001' AND subject='MOCAREV-0001'" \
  --work-group ambient-dev-analytics \
  --query-execution-context Database=ambient_dev_raw

05

Golden Parquet

AWS golden-combine Lambda · merges 3 zone files → 1 room file · powers ellamemory.com

Per-device (raw/)

DEV-MOH-301-ALiving Room

DEV-MOH-301-BBathroom

DEV-MOH-301-CEntry

14-column Schema V2 · no zone column · device partition in S3 path only

Golden (golden/)

golden_frames

14 columns + zone string
partition: date / facility / subject
UNION ALL of A + B + C

v_ambient_activity_index

SUM(ActivityCounts) per minute

v_zone_activity

Per-zone rows, filterable

Golden S3 Path

s3://ambient-dev-parquet-data/golden/date=2026-05-24/facility=FAC-MOCAREV-001/subject=MOCAREV-0001/ *.parquet

Athena UNLOAD writes multiple shards per subject. Partition projection on date/facility/subject — no device partition.

06

Athena Query Playground

Pre-built queries · live execution when AWS credentials configured

Date

SUM of AmbientActivityCounts per minute for one resident

sql

SELECT minute, AmbientActivityIndex, AmbientOccupancy, AmbientPosture
FROM ambient_dev_raw.v_ambient_activity_index
WHERE date='2026-05-24'
  AND facility='FAC-MOCAREV-001'
  AND subject='MOCAREV-0001'
ORDER BY minute;

Configure AWS_ACCESS_KEY_ID + AWS_SECRET_ACCESS_KEY in Vercel env settings to run queries live.

07

Cost Model

12 rooms · 36 devices · us-east-1 pricing · daily run cadence

Service	Formula	Per run	Per month (daily runs)
S3 PUT requests	36 files × $0.000005	<$0.001	$0.005
S3 Storage (raw+golden)	0.06 GB × 2 × $0.023	—	$0.081
Lambda golden-combine	600s × 256 MB × $0.0000167/GB-s	$0.003	$0.075
Athena scan	0.050 GB × $5/TB	<$0.001	$0.007
TOTAL		$0.003	$0.169/mo

08

Data Volume

At 1 fps sustained · us-east-1 pricing

Metric	Per device	Per room (3 devices)	Full pilot (36 devices)
Frames / day	86K	259K	3.11M
Raw JSON / day	13 MB	40 MB	0.5 GB
Gzip NDJSON / day	2 MB	6 MB	0.1 GB
Parquet+ZSTD / day	2 MB	5 MB	0.1 GB
Parquet / month	50 MB	0.1 GB	1.8 GB

09

CLI Reference

scripts/generate_sim_frames.py · ambientcloud repo

zsh

# Prerequisites
eval "$(./scripts/resolve_env.sh dev)"
export AWS_PROFILE=pilot AWS_DEFAULT_REGION=us-east-1

# Dry-run first
python3 scripts/generate_sim_frames.py --fps 18 --dry-run

# Full 12-room pilot at 18 fps
python3 scripts/generate_sim_frames.py --fps 18

# Specific rooms, custom date, scenario
python3 scripts/generate_sim_frames.py --rooms 301 302 303 --date 2026-05-19 --scenario fall

# Registry entries only (re-sync after DynamoDB changes)
python3 scripts/generate_sim_frames.py --registry-only

S3 Partition Path

s3://ambient-dev-parquet-data/raw/date=2026-05-24/facility=FAC-MOCAREV-001/subject=MOCAREV-0001/device=DEV-MOH-301-A/sim_frames.parquet

Partition columns (date / facility / subject / device) are path components only — not stored in the Parquet file body. Athena reads them via partition projection.

10

Frame Schema V2

15 data columns + 4 partition columns · ZSTD compression

Field	Type	Description
captured_at	timestamp(us, UTC)	UTC frame timestamp, microsecond precision
device_id	string	DEV-MOH-{room}-{A\|B\|C}
frame_number	int64	Monotonic counter from device boot
point_x	list<float32>	X coordinates of radar point cloud
point_y	list<float32>	Y coordinates (depth from sensor)
point_z	list<float32>	Z coordinates (height)
point_velocity	list<float32>	Radial velocity per point (m/s)
point_snr	list<float32>	Signal-to-noise ratio per point (dB)
track_ids	list<int32>	Tracker-assigned person IDs
track_max_z	list<float32>	Max height per tracked person
track_min_z	list<float32>	Min height per tracked person
AmbientActivityCounts	int32	Total radar points in this frame
AmbientOccupancy	int32	Persons tracked; 0=empty, 2+=visitor
AmbientPosture	string (nullable)	Standing · Seated · On Floor · null
Partition columns — path only, not in file body
date	string	e.g. 2026-05-19
facility	string	e.g. FAC-MOCAREV-001
subject	string	e.g. MOCAREV-0001
device	string	e.g. DEV-MOH-301-A

SimulationLab.

Simulation Console

Pilot Roster

Activity Preview

Device Configuration

How to Run

Golden Parquet

Athena Query Playground

Cost Model

Data Volume

CLI Reference

Frame Schema V2

Simulation
Lab.