How zackultima/ally works
Ally is presented as an Android companion and safety monitor, with local fall detection, consent-based expression tracking, conversation, and guardian escalation. The supplied tree primarily exposes Kotlin domain logic and Python model/data workflows rather than a complete Android integration, so the graph separates those branches. Fall-detection training and live inference are represented distinctly; companion and emotion tooling are data/model preparation components. Some product capabilities are documented but their runtime implementation is not present in the supplied tree.
Main entry points: FallFlow.kt Fall response · LevelRules.kt Mood check-ins · pipeline.py Live fall pipeline · data.py Window data
Components 28
Android domain logic
- Fall response
app/fall/flow/src/main/kotlin/ally/fall/flow/FallFlow.kt - Fall state
app/fall/flow/src/main/kotlin/ally/fall/flow/FlowState.kt - Fall screen
app/fall/flow/src/main/kotlin/ally/fall/flow/Screen.kt - Guardian rota
app/fall/flow/src/main/kotlin/ally/fall/flow/CallRota.kt - Mood check-ins
app/emotion/levels/src/main/kotlin/ally/emotion/levels/LevelRules.kt - Expression readings
app/emotion/levels/src/main/kotlin/ally/emotion/levels/Expression.kt - Mood log
app/emotion/levels/src/main/kotlin/ally/emotion/levels/MoodLog.kt - Phrase checking
app/speech/phrases/src/main/kotlin/ally/speech/phrases/PhraseChecker.kt
Fall detection
- Live monitor app
ml/fall/src/ally_fall/live/app.py - Live fall pipeline
ml/fall/src/ally_fall/live/pipeline.py - Pose keypoints
ml/fall/src/ally_fall/keypoints.py - Alert rule
ml/fall/src/ally_fall/live/alert.py - Labels and splits
ml/fall/src/ally_fall/omnifall.py - Dataset adapters
ml/fall/src/ally_fall/datasets - Window data
ml/fall/src/ally_fall/data.py - Fall models
ml/fall/src/ally_fall/models.py - Model training
ml/fall/src/ally_fall/train.py - INT8 export
ml/fall/src/ally_fall/export.py
Companion data
- Dialogue records
ml/companion/src/ally_companion/dialogue.py - Room context
ml/companion/src/ally_companion/roomnote.py - Held-out filter
ml/companion/src/ally_companion/heldout.py - Dialogue loaders
ml/companion/src/ally_companion/loaders.py
Emotion data
- Emotion datasets
ml/emotion/src/ally_emotion/utkface.py - Dataset conversion
ml/emotion/src/ally_emotion/convert_check.py
Other components
- Older adult
- Guardian
- Companion conversation
- Object recognition
Connections 25
- Older adult → Fall response: provides events
- Fall response → Fall state: updates state
app/fall/flow/src/main/kotlin/ally/fall/flow/FallFlow.kt - Fall screen → Fall state: reads state
app/fall/flow/src/main/kotlin/ally/fall/flow/Screen.kt - Fall response → Guardian rota: uses rota
app/fall/flow/src/main/kotlin/ally/fall/flow/FallFlow.kt - Fall response → Guardian: notifies
- Older adult → Mood check-ins: provides replies
- Mood check-ins → Expression readings: records readings
app/emotion/levels/src/main/kotlin/ally/emotion/levels/LevelRules.kt - Mood check-ins → Mood log: stores mood data
- Older adult → Phrase checking: provides phrases
- Live monitor app → Live fall pipeline: uses pipeline
ml/fall/src/ally_fall/live/app.py - Live fall pipeline → Pose keypoints: uses points
ml/fall/src/ally_fall/live/pipeline.py - Live fall pipeline → Alert rule: updates alert
ml/fall/src/ally_fall/live/pipeline.py - Labels and splits → Dataset adapters: provides identities
ml/fall/src/ally_fall/datasets/up_fall.py - Window data → Labels and splits: uses labels
- Window data → Dataset adapters: uses datasets
ml/fall/src/ally_fall/data.py - Model training → Window data: loads windows
ml/fall/src/ally_fall/train.py - Model training → Fall models: trains model
ml/fall/src/ally_fall/train.py - INT8 export → Fall models: exports model
ml/fall/src/ally_fall/export.py - Dialogue records → Room context: encodes context
ml/companion/src/ally_companion/dialogue.py - Dialogue loaders → Dialogue records: loads records
ml/companion/src/ally_companion/loaders.py - Dialogue loaders → Held-out filter: filters training
- Companion conversation → Dialogue records: trained from
- Emotion datasets → Dataset conversion: prepares data
- Older adult → Companion conversation: talks with
README.md - Older adult → Object recognition: provides view
README.md