Cognora+ – AI Companion for Cognitive & Emotional Wellness

Cognora+ is a GenAI-powered daily AI companion that detects emotional and cognitive decline from conversations and helps reconnect users with caregivers and loved ones.

Comment

✨ Inspiration

In today’s world, millions of elderly individuals live alone—physically safe, yet emotionally isolated and cognitively unsupported. Early signs of cognitive decline often go unnoticed until irreversible damage has occurred. Traditional check-ins by caregivers or family are irregular, subjective, and reactive.

Cognora+ was born out of this challenge—to create a proactive, intelligent, and emotionally safe AI companion that engages in daily personalized conversations, gently assesses cognitive health, and alerts caregivers before a crisis happens. Inspired by real-world stories of dementia in families and the loneliness epidemic among aging populations, Cognora+ blends the warmth of companionship with the precision of AI.

🤖 What it does

Cognora+ is an AI-powered wellness system designed for elderly users to safely interact, express, and be evaluated daily through short conversations. Here's how it works:

  • 🗣️ Personalized Conversation Companion
    Mii (the AI agent) speaks daily to the user, starting with light talk (e.g., weather, hobbies) and avoiding sensitive topics.

  • 🎤 Speech Capture & Analysis
    Voice input is transcribed using Amazon Transcribe and processed via Claude 3 Sonnet using custom prompts and agents (EmotionAgent, MemoryAgent).

  • 📊 Cognitive Health Indicators
    The conversation is analyzed for 10+ key indicators (e.g., fluency, lexical diversity, memory recall, emotional tone) to compute a Cognora Score (0–100).

  • 📈 Unique Baseline Tracking
    Unlike traditional models, users are only compared to their own historical data, building a personalized benchmark over weeks and months.

  • 🔔 Caregiver Alerts via SNS
    If scores drop significantly or daily check-ins are missed, caregivers are alerted via email/SMS (AWS SNS) to take early action.

  • 📊 Real-Time Dashboard & Reports
    A Streamlit dashboard visualizes the Cognora Score trend, emotional heatmap, and offers downloadable PDF reports.

  • ☁️ Cloud-Native Storage & Insights
    All data is stored securely in Amazon S3 and DynamoDB, and processed via Bedrock, following HIPAA-aligned principles.

🛠 How we built it

Cognora+ was engineered using a modular, cloud-native, and AI-first architecture:

  • Frontend:

    • Built with Streamlit and Plotly for real-time interactivity
    • Includes voice/text input interface, charts, alerts, and PDF export

  • AI Core (LLMs):

    • Powered by Claude 3 Sonnet via Amazon Bedrock
    • Multi-agent system using LangChain:
    • EmotionAgent → analyzes sentiment
    • MemoryAgent → evaluates coherence and recall
    • AlertAgent → triggers notifications

  • Backend Services:

    • Amazon Transcribe for speech-to-text
    • Amazon DynamoDB for storing user scores and metadata
    • Amazon S3 for transcripts, audio, and report files
    • AWS SNS for caregiver alerts
    • AWS Lambda for automated threshold checks

  • Infrastructure as Code (IaC):

    • Full infra defined via Terraform: reproducible, secure, and scalable
    • Includes automated provisioning of S3, DynamoDB, IAM, SNS, and Lambda

⚠️ Challenges we ran into

  1. Prompt Alignment:
    Designing prompt formats that consistently extract quantifiable data (not just summaries) from Claude Sonnet required deep experimentation.

  2. Multi-Day Memory & Trends:
    Streamlit is stateless. Creating user-specific, time-aware sessions meant building a DataManager layer to sync with DynamoDB efficiently.

  3. Emotion vs. Cognition Tradeoffs:
    Balancing emotional language detection with cognitive pattern analysis required tuning across Lexical Density, Fluency, and Affect Models.

  4. Elderly-Friendly Design:
    We prioritized ease of use, low-latency voice input, readable visualizations, and minimal tech friction for older adults.

  5. Non-Comparative Benchmarking:
    Building a model where users are compared to themselves only (and not others) required a dynamic baseline computation model.

🏆 Accomplishments that we're proud of

  • ✅ Built a production-ready GenAI product, not just a prototype
  • 🧠 Developed a scientifically grounded, personalized scoring system with real clinical relevance
  • 📣 Created a fully automated caregiver safety loop
  • 🔧 Integrated 10+ AWS services using Terraform & Boto3
  • 📊 Designed a clean, real-time dashboard accessible to both seniors and families
  • 💡 Engineered an emotionally aware, human-first AI companion

📚 What we learned

  • How to operationalize GenAI responsibly in the health + aging domain
  • Mastery of Claude Sonnet prompt engineering for structured outputs
  • The importance of ethical AI practices for vulnerable populations
  • The power of LangChain’s modular agents in chaining reasoning tasks
  • How to create a zero-friction user experience for elderly and non-tech-savvy users
  • Terraform and AWS best practices for scalable infra deployment

🚀 What's next for Cognora+ – AI Companion for Cognitive & Emotional Wellness

We are committed to transforming Cognora+ into a real-world product and scaling it with the following roadmap:

  • 🔁 Long-Term Memory: Use vector stores to track changes across months
  • 🧘 Personalized Coping Recommendations: Mindfulness tips, music, or safe conversation topics
  • 🧠 Voice Biomarker Integration: Detect speech rate, tremor, tone, and hesitation
  • 🔐 Caregiver/Clinician Portal: Multi-user access to dashboards and trends
  • 🩺 Clinical Trials & Partnerships: Collaborate with geriatric research groups for score validation
  • 🌐 Localization: Japanese, Spanish, and Hindi language support
  • 📱 Cross-Platform App: Web, mobile, and Alexa integration

Cognora+ bridges AI with empathy. It doesn’t just analyze — it listens, learns, and acts when it matters most.
It’s not just a hackathon project — it’s a movement toward empathetic AI for aging with dignity. 💛

Built With

Share this project:

Updates