Memory Lane

Memory Lane recreates loved ones as AI companions and organizes shared photo galleries to support memory recovery, emotional bonding, and patient well-being.

Comment

🧠 Memory Lane

Built with: SvelteKit · Node.js · Fastify · Prisma · Cloudinary · Conversational AI · LLM APIs

Inspiration

Every 3 seconds, someone in the world develops dementia. Over 55 million people already live with memory disorders — projected to reach 78 million by 2030 and 139 million by 2050. Behind each case is a family struggling to maintain emotional and cognitive connection.

Caregivers spend 20–60+ hours per week repeating the same grounding facts (names, places, routines). Traditional “memory aids” — static journals or photo albums — fail to adapt to emotional states or reinforce memory dynamically.

We built Memory Lane to transform those passive artifacts into living, conversational companions that preserve identity, reduce caregiver fatigue, and create moments of clarity.

What it does

Memory Lane is an AI reminiscence and identity-preservation platform.

It ingests curated patient context — photos, short audio clips, life milestones, relationship graphs, and routines — and uses conversational AI to instantiate personalized replicas of loved ones or familiar personas.

Patients chat naturally (text or voice) with these replicas, who recall and discuss memories tied to gallery items. Each session is logged for caregivers with metrics like duration, topic depth, sentiment, and cognitive coherence — providing insights into emotional and cognitive stability.

🌸 Key Features

  • AI Family Replicas: Familiar personas that reinforce memory and reduce confusion.
  • Interactive Galleries: Secure, curated media accessible through a calming interface.
  • Guided Recall Sessions: Structured reminiscence exercises led by replicas.
  • Caregiver Control: Full oversight of patient access and data.
  • Adaptive Recall Patterns: Prompts evolve based on engagement and recall success.

How we built it

⚙️ Architecture

Frontend (SvelteKit):

  • Components: ChatWindow.svelte, Gallery.svelte, ReplicaWizard.svelte, AnalyticsDashboard.svelte
  • Accessibility-first: large tap targets, reduced cognitive load, high-contrast mode

Backend (Node.js / Fastify / Prisma):

  • Patient & caregiver models with secure linkage
  • AI integration layer for replica management and chat
  • Cloudinary-based media ingestion and metadata normalization
  • Session logging pipeline for analytics

AI Layer:

  • Personalized replica creation from curated data
  • Knowledge base ingestion (text + captioned photos)
  • Contextual recall using conversational retrieval
  • Per-user scoping and session telemetry for analytics

Challenges we ran into

  • Ethical Design: Avoiding false intimacy and ensuring replicas never impersonate real people.
  • Context Quality: Balancing multimodal inputs (text, image, audio) with fast inference.
  • Adaptive Memory Recall: Building recall patterns that evolve with engagement history.
  • Quantifying Emotion: Translating qualitative memory engagement into measurable cognitive metrics.
  • Safety & Privacy: Encrypting data while maintaining fast, seamless caregiver workflows.

Accomplishments that we're proud of

📊 Quantifiable Impact

Metric Baseline Target / Observed
Caregiver Onboarding Time ~30–40 mins (manual journaling) <5 mins via guided wizard
Average Session Length 5–8 mins 12–18 mins sustained attention
Repetitive Grounding Tasks Offloaded 0% 35–45% handled by AI
Patient Confusion Recovery Time ~2–3 mins <30 sec via contextual recall
Photo Recall Success Rate 40–50% 70–80% using AI-guided recall
Mood Stability (sentiment tracking) Fluctuating Consistent improvement after 2 weeks
Caregiver Stress Index High baseline ~30% reduction after use
Patient Engagement Consistency Low adherence +60% increase in weekly sessions
Session Data Utilization None 100% logs stored & visualized

What we learned

  • AI empathy works best under supervision. Memory Lane showed that human-guided AI can support reminiscence therapy without replacing caregivers.
  • Engagement metrics are powerful clinical signals. Mood stability and recall latency can act as cognitive health indicators.
  • UX simplicity equals adoption. A 5-minute onboarding lowered caregiver friction dramatically.
  • Guardrails matter. Transparency and caregiver-controlled replicas prevented ethical and emotional risks.

What's next for Memory Lane

We’re extending Memory Lane into a more clinically aware tool and research platform.

🔮 Roadmap

  • Emotion-aware tone modulation – adaptive empathy via sentiment detection
  • Vector-based multimodal embeddings – correlating image + text recall patterns
  • AI-assisted media captioning (with caregiver approval)
  • Cognitive trend visualization – long-term recall tracking dashboard
  • Secure clinician tokens – allow doctors or therapists to review anonymized engagement logs

🧠 Future Research Potential

Memory Lane’s session data can inform early cognitive decline detection by:

  • Tracking recall latency across weeks
  • Analyzing topic coherence patterns
  • Measuring longitudinal consistency of autobiographical responses

🛡️ Ethical Framework

  • No false memories or impersonations
  • Soft correction strategies (“Let’s look at this memory together…”)
  • Caregiver-supervised AI knowledge bases
  • Privacy-first: encrypted storage, minimal identifiable data

Accomplishments that we're proud of (Human Impact)

  • Turned static photos and notes into dynamic memory sessions
  • Designed a non-intrusive, emotionally grounded AI interface
  • Enabled families to reconnect through memory reinforcement
  • Provided new clinical metrics for understanding memory progression

Closing Statement

Memory Lane reimagines memory care as a bridge between technology and empathy.
By turning static life artifacts into dynamic, identity-reinforcing conversations, it helps patients remember who they are — one story at a time.

Github link: https://github.com/AlphaTechini/Built-with-Sensay-API

Built With

Share this project:

Updates