How We Built an AI Facilitator That Helps Families Make the Hardest Decisions Together

The Challenge

When a parent's health declines, families face a cascade of urgent decisions: Should Dad move to assisted living? Who manages the finances? Which doctor should handle the new diagnosis? These conversations happen over scattered phone calls, group texts, and tense holiday dinners — if they happen at all.

The problems compound:

Geographic scatter — siblings live in different states and time zones, making synchronous discussion nearly impossible
Information asymmetry — the sibling who lives closest has different context than the one who manages finances, creating conflict rooted in incomplete pictures
Emotional overload — care decisions carry enormous emotional weight, and unstructured conversations quickly devolve into old family dynamics
No paper trail — decisions get made (or not made) with no record of who agreed to what, leading to repeated arguments
Analysis paralysis — families with multiple options and no framework for evaluation get stuck, sometimes for months, while care needs escalate

Existing tools — group chats, shared docs, video calls — weren't designed for structured decision-making. Families needed something purpose-built: a platform that could facilitate difficult conversations, ensure every voice was heard, and guide groups toward decisions they could all support.

Our Solution

We built AllWise — a collaborative platform where families make care decisions together, guided by AI that facilitates rather than dictates. The platform is live at beallwise.com.

Family Workspaces

Each family gets a private workspace with role-based access (owner, admin, member). Family members join via email invitation with secure token-based onboarding. The workspace organizes everything around decisions — specific topics the family needs to resolve.

Decision Rooms

Each decision gets its own real-time chat room powered by Socket.io. But these aren't just chat threads — the AI operates as an invisible facilitator working in three layers:

Intent Extraction — As family members share their thoughts, the AI identifies underlying needs, concerns, and priorities. "I'm worried about Dad being alone" becomes a structured intent: safety concern, high importance, emotional weight.
Conflict Detection — When messages reveal tension or opposing views, the AI detects the conflict pattern and can intervene with reframing questions or request clarification.
Consensus Building — The AI generates proposals based on extracted intents, evaluates alignment scores between proposals and each family member's priorities, and tracks the group's consensus state through discussion stages.

Structured Decision Tools

Beyond chat, families have access to anonymous and identified polls (with auto-expiration and tie-breaking), proposal evaluation cards that show how well each option addresses each family member's concerns, and a document vault where financial, legal, and healthcare documents can be shared, auto-categorized, and analyzed by AI.

AI Facilitation Levels

Families can tune the AI's involvement: low (mostly observes, surfaces intents), medium (suggests questions, generates proposals), or high (actively facilitates, detects stalls, recommends next steps). The AI is a facilitator, not a decision-maker — the family always decides.

AllWise platform features - team collaboration, intelligent guidance, security, and organization

Our Approach

We built AllWise as a production-grade Next.js application with a sophisticated real-time backend, designed for the emotional sensitivity and privacy requirements of family care coordination.

Data Architecture

The PostgreSQL schema spans 56 Prisma models covering the full lifecycle of collaborative decision-making: Users, Families, FamilyMembers, Decisions, Messages, Intents, Proposals, ProposalEvaluations, ConsensusStates, Polls, PollVotes, AIConversations, Documents, and more. The schema enforces referential integrity across all relationships while supporting the flexible nature of family discussions.

Real-Time Infrastructure

We implemented a dual real-time architecture: Socket.io on the application server for low-latency message delivery, plus AWS Lambda WebSocket handlers (connect, disconnect, message) backed by DynamoDB for connection tracking. This ensures reliable real-time updates even when family members are on different devices and networks.

AI Pipeline

The AI facilitation system operates as a pipeline:

Message intake — Every message is analyzed for intent signals, emotional tone, and conflict markers
Intent extraction — Structured intents are extracted with importance rankings and sentiment analysis
Facilitation triggers — Configurable rules determine when the AI should intervene (based on facilitation level, stall detection, conflict escalation)
Proposal generation — The AI synthesizes intents into concrete proposals with alignment scores
Consensus tracking — The system monitors group alignment through discussion stages toward resolution

Security & Privacy

Family care discussions contain deeply personal information. Authentication uses JWT + Iron Session with httpOnly cookies. Email verification is required for all accounts. Invitation tokens are single-use and time-limited. Document access is controlled per-family with S3 pre-signed URLs for secure file delivery. All sensitive data transmission uses encryption in transit.

Email & Communication

Transactional emails (invitations, verification, password reset, meeting reminders) use Resend API with custom branded templates, backed by cross-account AWS SES for reliable delivery from the beallwise.com domain.

Results & Outcomes

AllWise launched at beallwise.com as a production-ready platform with a complete feature set for family care decision-making.

Platform Capabilities

Full family workspace lifecycle — creation, invitation, role management, and settings — with Stripe subscription billing
Real-time decision rooms with WebSocket-powered messaging, file sharing, and AI facilitation
Three-layer AI system — intent extraction, conflict detection, and consensus building — all configurable per family's comfort level
Structured voting with anonymous polls, auto-expiration, reminders, and tie-breaking
Document vault with hierarchical folders, auto-categorization (financial, legal, healthcare), visibility controls, and AI-powered text extraction and summarization
Private AI chat — each family member can have separate conversations with the AI to process their thoughts before sharing with the group
Calendar integration for scheduling family meetings with timezone-aware reminders
Care resources library with searchable articles and guides

Technical Delivery

56 data models in a PostgreSQL schema that captures the full complexity of collaborative decision-making
Dual real-time architecture — Socket.io + AWS Lambda WebSockets with DynamoDB connection tracking
Model-agnostic AI — OpenAI GPT-4o primary with AWS Bedrock (Claude) ready as an alternative
E2E test coverage with Playwright across authentication, decision, and collaboration flows
AWS infrastructure — Amplify hosting, RDS PostgreSQL, S3 storage, Lambda functions, DynamoDB, SES email

AllWise demonstrates that AI can enhance human decision-making without replacing it. By extracting structure from emotional conversations, making invisible priorities visible, and guiding groups through a proven facilitation framework, the platform helps families navigate the hardest decisions they'll ever face — together.