How the pieces fit together
Specialized prediction engines, an orchestrator, local AI explanations, and backend neural-network training support one calm daily view. Track what happened, understand the available estimate, and plan your next step with context you can revisit.
The prediction stack
Prediction values, coordination, explanations, and training are related, but they are not the same thing.
Registered engines focus on distinct sleep targets: the next sleep opportunity, bedtime, and waking from an active sleep. They use the recorded history and current routine context that the application makes available. A result can be unavailable when the history does not support a useful estimate.
An orchestrator coordinates those target-specific results into the daily view. It keeps recorded events, available estimates, and missing guidance distinct so one target does not quietly stand in for another. You see a planning aid that reflects what is currently available.
DreamMorrow can use a privately hosted Ollama service to explain verified prediction context in everyday language. “Local” means the service runs on our infrastructure; it does not mean an on-phone or offline model. This layer explains context and does not set times or replace a prediction engine. It is different from a general parenting chatbot.
The backend also supports PyTorch neural-network training and evaluation. Candidate models are trained and assessed as research and engineering work, then kept distinct from the caregiver models that are activated for use. Training capability does not mean a neural model is running for every child or every prediction.
Designed for real days
The useful part is a record you can trust enough to revisit, share, and edit.
Editable logs: correct a missed timer or add a sleep later, because a family’s day rarely fits a perfect sequence. The history remains something you can review rather than a score you have to protect.
Authorized sharing: invite another caregiver to the child profile so everyone works from the same record, with access controlled through the account.
Connected use: DreamMorrow needs an internet connection to load and save current records. A Home Screen installation does not make sleep tracking an offline workflow.
Estimates are planning context, not medical advice or a promise. They can be absent when the available history is not enough, and real life can always change what happens next.
A plain-language answer
An explanation gives the estimate a useful place in the day without turning it into a promise.
When the relevant context is available, an explanation can name the recorded anchor, describe recent comparable sleep context, and make the flexibility of a planning window easier to understand. It keeps the estimate connected to the history you can review.
Early wakes and short naps can be part of the recorded context when the supported prediction path uses them. The explanation should describe what is present in the record, not invent a cause or claim that every event shifts a time.
Illustrative wording, not live child output: “This estimate draws on the latest completed nap and recent comparable days. Use the window as a flexible guide.”
Keep exploring
Read the DreamMorrow sleep app comparison for a published-feature view of Napper and Huckleberry, or return to the Dreamline preview for a sample day.
