Technical Architecture & AI Strategy

Production-Grade Conversational Architecture

MemorableLife is a "Therapeutic AI" companion designed to alleviate loneliness and mitigate cognitive decline in seniors. Unlike standard chatbots, our architecture is built to deliver Reminiscence Therapy—a clinical intervention that requires deep, persistent memory, empathetic pacing, and culturally specific context.

To achieve this we have transitioned from our existing Minimal Viable Product (MVP) to a more robust, Production-Grade Conversational Architecture. This significant technical evolution is driven by a newly defined production schedule aimed at launching a pilot release in January-February 2026.

The upgraded architecture is designed to directly address critical limitations identified in the prototype. By implementing advanced capabilities such as a vector-search-enabled "Biographer" engine for long-term memory and a "Continuous Listening" interface for natural dialogue, we are building a platform capable of delivering a truly therapeutic and seamless experience for our senior users. This phased development plan moves beyond simple data logging to creating a system that understands and retains the "searchable meanings" of a user's life story.

1. The MemorableLife Neural Biographer Engine™

At the core of MemorableLife is our proprietary 'Neural Biographer Engine', an event-driven AI architecture designed to capture the nuance of human memory. Unlike traditional databases that rely on rigid keywords, our system utilises Semantic Vector Search to understand the meaning and context behind a senior's stories. Designed by speech and reminiscence therapists, MemorableLife transforms raw AI into a Supportive Witness.

Our Architecture

We are proudly built 100% on Google Cloud Platform, leveraging a serverless, scalable infrastructure:

Ingestion Tier: We utilise Cloud Functions (2nd Gen) to process voice and text inputs in real-time.

Cognitive Processing: User inputs are analysed by Vertex AI (Gemini), using the text-embedding-004 model to convert complex narratives into 768-dimensional vector embeddings.

Neural Storage: Memories are stored in Google Firestore, utilising its native Vector Search capabilities. This allows our AI to perform "Nearest Neighbour" retrieval, instantly connecting a question like "What did I do in the summer?" to a memory about "planting roses in 1970"—even if the keywords don't match.

2. The User Experience

Voice-First Accessibility

Recognising that seniors often struggle with "wake words" and button presses, we are pioneering a Continuous Listening interface.

Fluid Dialogue: Leveraging Vertex AI to orchestrate Gemini models and Voice Activity Detection (VAD), MemorableLife enables a "Walkie-Talkie-Free" experience where users can speak naturally, pause to think, and be interrupted gently, mirroring human conversation.

Hyper-Realistic Audio: We utilize Google's Journey (Generative) TTS voices to deliver a warm, empathetic tone that is indistinguishable from a human carer—critical for building trust with users facing dementia.

3. Safety and Trust

The "Ethical Wall": Two-Tier Privacy Architecture

Trust is our currency. We have engineered a proprietary Two-Tier Privacy Filter that goes beyond standard encryption to act as a "Clinical Governor" for all interactions.

Tier 1: Real-time Content Filter (Data Routing & Safety)
- Real-Time Triage: Before storage, a specialized model classifies every user input into one of three security levels: Public, Restricted, or Distress.

The Dual-Channel Alert: If the classifier detects a "Distress" signal (e.g., severe anxiety, self-harm risk), it triggers an immediate, parallel notification system via SMS and Email to designated guardians.

Tier 2: The Clinical & Confidentiality Safety Net
- The "Confessional" Firewall: We prioritise the user's dignity. Highly sensitive disclosures classified as "Restricted" are sequestered from memory summaries and the Caregiver Dashboard. This ensures seniors can speak freely to their AI companion without fear of surveillance.

Clinical Guardrails: The system actively scans outgoing AI responses for medical advice or diagnostic language. If detected, the response is intercepted and replaced with a clinically approved safety phrase (e.g., "You should seek medical advice, I am not a doctor..."), ensuring the AI never oversteps its therapeutic boundaries.

Conversational Ring-Fencing: The ethical instruction in the Gemini System Prompt that forbids the AI from discussing controversial topics and forces it to focus on therapeutic, nostalgic "cultural touchstones."

Google Pixel 10 displaying MemorableLife app.

Medical-Grade Privacy by Design

MemorableLife is built on the global backbone of Google Cloud Vertex AI. We protect our users with a multi-layer security architecture:

Identity Isolation: Every senior’s memories are stored in a private, encrypted vault, accessible only to them and their chosen loved ones.
Service Account Authorization: Our mobile apps use "Digital Passports" (Service Accounts) to talk to the AI brain, ensuring no personal keys are ever stored on the phone.
Real-Time Safety Filters: Our AI continuously monitors for signs of distress or health alerts, providing instant triage and notifications to human caregivers.