# App Research Final Decision

Status: Complete with Firecrawl and Tavily evidence.

Last updated: 2026-05-07

## Decision

Build the app.

Build it narrowly as:

> A Thai-first AI caregiver wellness co-pilot for families caring for elderly parents with Alzheimer's disease or dementia at home.

Do not build it as:

- A patient memory-recording app.
- A dementia diagnosis app.
- A cognitive screening app.
- An AI companion robot or facility platform.
- A caregiver marketplace.
- A generic mental-health chatbot.

## Why This Decision Is Strong

The Firecrawl and Tavily competitor pass found many adjacent products, but no clear product with this exact combination:

- Thai-first.
- Dementia-specific.
- Caregiver-first.
- Voice-first.
- Burden-aware.
- Structured-output based.
- Family-handoff oriented.
- Non-diagnostic and safety-bounded.

This is the right hackathon scope because it is differentiated, demoable, useful, and safer than diagnosis or treatment.

## What To Build In MVP

### 1. Thai caregiver check-in

Input can be Thai text for the first prototype, with voice as the high-impact demo layer if time allows.

The check-in prompt should ask:

- What happened?
- How hard was it for you?
- Is there any immediate danger?

### 2. Structured incident log

Every check-in should produce structured output:

- Incident type.
- Short summary.
- Possible trigger.
- Patient behavior.
- Caregiver feeling.
- Burden score.
- Risk level.
- Safe next step.
- Family help request.
- Follow-up reminder.

### 3. Safe next step

The response should be practical and bounded.

Examples:

- Try a calm repeated-answer card for repetitive questions.
- Remove immediate hazards and call family if wandering risk appears.
- Contact a clinician if there is sudden confusion, injury, medication concern, or major behavior change.
- Call emergency services for immediate danger.

### 4. Family handoff message

Generate a short message the caregiver can review before sharing.

The message should be:

- Calm.
- Specific.
- Non-blaming.
- Action-oriented.
- Thai-family appropriate.

### 5. Weekly insight view

Show simple trends:

- Burden score over time.
- Most common incident types.
- High-risk count.
- Common triggers.
- Help requested.
- Notes for doctor or family meeting.

## What To Copy

| Competitor pattern | Copy this | Do not copy this |
|---|---|---|
| Memory Aid AI | Low-friction capture and recall | Always-on or unclear transcript privacy |
| Alzheimer's Care Partner | Care tracking and family team framing | Overbroad AI advice without clear local boundaries |
| Elevmi | Observation recording and doctor-prep workflow | Non-Thai positioning and unvalidated secondary-source claims |
| Amicus Brain / RAZ Care | Conversational dementia-care AI advisor | Device/ecosystem dependency |
| Dementia CareAssist | Practical behavior-specific guidance | Static-only content workflow |
| CogniCare | Dementia carer companion concept | Unclear availability and non-Thai positioning |
| IanaCare / CaringBridge / Lotsa Helping Hands | Family support and help-request workflow | Manual coordination burden |
| Wysa / Woebot | Safety boundaries and responsible AI language | Generic wellness bot experience |
| ElderThai / OnCare / ThaiHelper | Thai home-care practicality and role boundaries | Marketplace-first product strategy |
| Dinsaw / AIT Elder Care AI / CloudNurse | Proof that Thai elder-tech and AI elder-care activity exists | Robotics, research-platform, or facility-admin drift |
| BrainCheck / Neurotrack | Professional-use caution | Screening, diagnosis, or clinical scoring |

## What To Avoid

### Avoid diagnosis

Never say the app detects, diagnoses, predicts, treats, or monitors dementia clinically.

### Avoid medication advice

The app can help the caregiver remember to discuss medication issues with a professional. It should not recommend changing medicine.

### Avoid hidden recording

No always-on capture. No invisible family sharing. No patient surveillance framing.

### Avoid generic chatbot positioning

The app should generate useful care artifacts, not just supportive conversation.

### Avoid building a marketplace

Service referral can be a fallback. It should not become the main product in the hackathon.

### Avoid clinical dashboard styling

Use plain caregiver language. The visualization should help family understanding, not imply medical diagnosis.

## Product Name Direction

Working internal description:

> Thai Caregiver Wellness AI

Pitch phrase:

> Voice in, care memory out.

Alternative pitch phrase:

> A second brain for the caregiver, not surveillance for the patient.

## Demo Script

1. Caregiver records a short Thai check-in after a difficult care moment.
2. App transcribes or accepts text.
3. AI produces a structured incident log.
4. App shows burden and risk level.
5. App suggests one safe next step.
6. App drafts a family handoff message.
7. Dashboard shows weekly stress and incident patterns.

## Data Visualization Plan

Show these charts:

| Visualization | Data | Insight |
|---|---|---|
| Burden trend line | Daily burden score | Shows caregiver strain becoming visible |
| Incident type bars | Count by incident type | Shows what kind of care problem repeats |
| Risk badge summary | Low/medium/high count | Shows when escalation may be needed |
| Trigger chips | Extracted trigger frequency | Helps family adjust routines |
| Family help tracker | Help requested vs completed | Shows whether caregiver is supported |
| Clinician note panel | Weekly structured summary | Helps prepare for doctor or nurse discussion |

## Build / Copy / Avoid

### Build

- Thai caregiver voice/text check-in.
- Structured incident extraction.
- Burden score.
- Safe next step.
- Family handoff.
- Weekly insight visualization.

### Copy

- Care-team framing from Alzheimer's Care Partner and family coordination apps.
- Safety language from AI wellness and clinical products.
- Local practicality from Thai elder-care services.
- Behavior guidance structure from dementia caregiver tools.

### Avoid

- Diagnosis.
- Treatment advice.
- Patient surveillance.
- Generic chatbot UX.
- Marketplace-first scope.
- Clinical scoring.

## Final Recommendation

Continue with the current app direction and start implementation from the smallest useful workflow:

> Thai text/voice check-in -> structured incident log -> burden signal -> safe next step -> editable family handoff -> simple weekly insight.

This is the clearest app opportunity found in the competitor research.