# Team Decrypton

- **Event:** [Meta Global AI Developer Hackathon](https://cerebralvalley.ai/e/meta-ai-developer-hackathon)
- **When:** Oct 26, 12:00 PM – Nov 4, 9:00 PM (EST)
- **Where:** Online
- **Team:** [Sumesh Chodankar](https://cerebralvalley.ai/u/sumesh99)
- **GitHub:** https://github.com/SumeshChodankar/Anchorcare
- **Demo video:** https://youtu.be/wURSUnDup5M?si=AFyGUbPFWN4VYLrX
- **Gallery:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4278

AnchorCare + Engram

Tagline: A memory companion for people living with dementia that only says what a caregiver has approved, and shows its evidence every time.

The Problem

For someone with dementia, simple questions like “Who’s visiting today?” or “Did I take my medication?” can trigger anxiety and be repeated many times a day. Typical AI assistants aren’t suitable because they can invent information, which creates confusion and distress.

The Solution

AnchorCare + Engram is an AI companion that never invents memories.

Caregivers build a trusted Memory Ledger of approved facts, routines, medications, photos and family information. Nothing becomes visible to the patient without approval, and every change is fully audited.

Patients use a simple Companion that answers questions using only approved records. Answers are generated with Meta Llama 4 and always cite the memory records behind them. If there’s no evidence, the Companion says so rather than guessing.

When a patient needs help, one tap sends an SOS. Through Twilio, it rings a caregiver’s real phone, wherever they are.

Key Features

✅ Caregiver-approved memory records
✅ Evidence-backed AI answers with citations
✅ Medication and routine reminders
✅ Familiar-face recall from approved photos
✅ SOS alerts plus real phone calls to caregivers via Twilio
✅ Complete audit trail for every memory update

Real-World Use Cases

Who is this? Help the patient recall family and friends from approved photos.
Reassurance on repeat. Give the same calm, accurate answer every time a question is asked.
Medication support. Timed reminders, with patient responses reviewed by caregivers.
Care-team handoffs. Shared visibility across shifts and caregivers.
Remote family involvement. Loved ones contribute memories for approval, and can be called from anywhere.
Distress support. One-tap SOS alerts paired devices and phones a caregiver, even in another city.

Why Meta Llama?

Llama is open-weight, so the same models can run through Meta’s hosted API or on infrastructure a care provider controls. Smaller Llama models can even run on ordinary local hardware with Ollama. This enables:

better privacy
offline resilience
low operating costs
future multilingual responses, safety screening and automated care summaries

Built With

Meta Llama 4 · MongoDB Atlas · Twilio · React + TypeScript · Express

What Makes It Different?

AnchorCare’s core promise is simple: every answer is backed by caregiver-approved evidence. If the system doesn’t know, it says so. It never invents a memory.

Built entirely using synthetic patient data. AnchorCare is a support tool, not a medical device, and does not provide clinical advice.

## More from Meta Global AI Developer Hackathon

- [Nur : ReproCam](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4275)
- [DischargeBuddy](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4276)
- [Siterra](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4277)
- [Explanimate](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4279)
- [Sonor](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4280)
- [H-Meta](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4281)

---

Markdown version of https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4278. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt