Project DANDA

DANDA #001 — CareLoop: An AI Chief of Staff for the 63 Million Americans Caring for Someone They Love

AR Akhil Reddy Danda · 17th July, 2026 · 4 min read
DANDA #001 — CareLoop: An AI Chief of Staff for the 63 Million Americans Caring for Someone They Love

Project DANDA is my daily exercise in founder thinking: one real problem, one buildable software company, full architecture. Idea #001.

The problem

There are now 63 million Americans taking care of a family member — one in four adults. That number grew 45% in a decade, and it's still climbing as the population ages faster than the care workforce grows.

Here's what that actually looks like: an adult child with a full-time job spending 25 hours a week — a second, unpaid job — on someone else's life logistics. Booking specialists and fighting for earlier slots. Refilling nine prescriptions across three pharmacies. Appealing an insurance denial with a 40-minute hold time. Texting siblings the same update three times. Twenty-nine percent of these people are 'sandwich generation' — raising kids while doing all of this. Seventy percent say caregiving is damaging their careers. The economy quietly eats $600 billion a year in their lost wages.

The cruel part: almost none of those 25 hours is care. It's coordination. Paperwork. Phone trees. The loving part of caregiving is being crowded out by the administrative part — and the administrative part is exactly what software agents just became good at.

The idea: CareLoop

An AI chief of staff for the family caregiver. You connect the inputs once — the care recipient's appointment emails, patient portals, pharmacy accounts, insurance plan, the family group chat. CareLoop's agents then run the operations continuously: they draft the refill requests, find and propose appointment slots that fit your calendar, pre-fill the insurance appeal with the plan's own policy language, book the ride to dialysis, and post a weekly plain-English update to the family. Nothing executes without a human tap. Every action arrives as a card on your phone: approve, edit, or decline. You go from 25 hours of doing to 20 minutes of deciding.

Architecture

CareLoop — Agentic Care Operations PlatformINPUTSPatient portals · EmailPharmacy · Insurance EOBsFamily chat · Voice notesINGESTIONParsers · OCR · STTEntity extractionDedup + PHI redactionCARE GRAPHMeds · Providers · PlansAppointments · History(the long-term memory)AGENT ORCHESTRATORPlanner (LLM) → task graph → specialist agents:Scheduler · Refill · Claims/Appeals · Transport · CommsEach with tools, guardrails, and cost budgetsHUMAN GATEApprove / edit / declineevery action (mobile)Full audit trailACTION LAYERCalendar + portal booking APIs · e-refill · payer FHIR/queues · ride booking · voice-agent phone callsWeekly family digest · Everything logged, reversible, and encrypted (HIPAA posture from day one)

Why an agent loop, not an app

Caregiving admin is recurring, interrupt-driven, and cross-system — the exact shape of work where a to-do app fails and an agent shines. The planner decomposes each incoming event ('EOB says claim denied') into a task graph; specialist agents execute with narrow tools; the human gate keeps trust and liability sane. The Care Graph is the moat: after three months, CareLoop knows the care situation better than any human could document it.

Build plan (90 days)

Wedge: prescription refills + appointment scheduling only, for one metro, integrated with the two biggest portal vendors. Stack: FastAPI + Postgres + a task-graph orchestrator; LLM planner with function-calling; mobile approval PWA; SOC2/HIPAA posture from day one. Pricing: $29/mo family plan; later, employers buy it as a benefit — 70% of caregivers are struggling at work, and employers are already paying for that in attrition.

Why now

Voice agents can finally survive a pharmacy phone tree. Payers are being pushed toward FHIR APIs. And the demographic curve only bends one way. Somebody is going to build the operating system for family care — the only question is who gets the Care Graph first.

in Share on LinkedIn 𝕏 Post
Sources I read for this:
← More from Reddy Pulse