Applesandbanan.ai
Private AI parenting assistant for couples that tracks baby facts, open questions, and decisions across conversations
Side·May 2026 — Present·Solo
Problem
- New parents are sleep-deprived information processors. Pediatrician advice, sleep patterns, feeding amounts, and one-off observations arrive faster than they can be retained, and most of it lives in one parent's head while the other is asleep.
- Generic chat assistants forget everything between sessions, so each conversation starts from zero: name, age, history, prior decisions all have to be re-explained.
- Existing parenting apps are tracking-first (log a feed, log a nap) and don't actually help the couple think through the open questions they're working on together.
- Couples fall out of sync, hearing different doctors and trying different things, without a shared place that remembers what 'we' have decided.
Approach
- Built a family-scoped workspace where both parents share a single AI context: baby profile, accumulated facts, open questions, and prior decisions persist across every conversation.
- Designed an explicit memory model with three kinds of records: facts (e.g. 'sleeps in a Halo swaddle'), open questions ('starting solids next month?'), and decisions ('cardiology follow-up at 6 months'). The assistant's recall is structured rather than vibes-based.
- Single-tenant-per-family architecture: each family's data is private, with auth-gated access and granular write permissions for partner accounts.
- Shipped on a deliberately boring stack (one GCE VM, Docker Compose, SQLite on a persistent disk, Caddy for TLS), sized for family-scale rather than a SaaS pitch deck.
- Operational maturity baked in: Terraform-managed infra, nightly SQLite-to-GCS backups, Workload Identity Federation for keyless CI, structured JSON logs into Cloud Logging, Resend transactional email.
- Zero-touch deploys: push to 'main' ships build, DB migration, and container restart in under five minutes.
Tech Stack
Frontend
ReactViteTypeScriptTailwindTanStack Query
Backend
PythonFastAPISQLAlchemyAlembic
Data & Storage
SQLite
Infrastructure
GCPGCEDocker ComposeCaddyTerraformSecret ManagerArtifact RegistryCloud Logging
Testing
pytestvitest
Other
Anthropic APIResendGitHub ActionsWorkload Identity Federation
Outcomes
- Running in production with a family using it daily.
- AI conversations stay in sync between partners, with no 're-explaining the baby' each session.
- Single-VM topology keeps monthly cost under the GCP free tier while still supporting structured logging, monitoring, and nightly backups.