AWS Builder Center
CogniTrace: A Voice-First AI Companion for Cognitive Care for Dementia Patients

CogniTrace: A Voice-First AI Companion for Cognitive Care for Dementia Patients

According to the WHO, over 57 million people battle dementia globally, with 10 million new cases yearly—leaving families trapped between clinical visits and exhausting daily care. CogniTrace bridges this critical therapy gap through an intuitive, voice-first AI companion. Designed with a zero-friction UI for older adults, it seamlessly blends routine guidance with personalized reminiscence therapy and continuous cognitive support.







CogniTrace: A Voice-First AI Companion for Cognitive Care using AWS
Team: Prometheus Team Members: Nikhil Verma (@nickhilverma), Garima Varma (@garimavarma), Mohit Srivastava (@mohitsr), Shreyash Raj (@shreyash_raj) App Category: Social Impact, Health Care



Our Vision

What if the most impactful cognitive support didn't require a clinical setting—but unfolded naturally through everyday dialogue at home?
According to the WHO, over 57 million individuals lived with dementia in 2021, accompanied by nearly 10 million newly diagnosed cases annually—with upwards of 60% residing in low- and middle-income nations. While prevalence peaks past age 65, young-onset dementia still represents up to 9% of all diagnoses.
Yet a critical blind spot persists: patients navigating early-stage Alzheimer’s disease and Mild Cognitive Impairment (MCI) actively attend speech therapy and memory clinics, only to encounter a silent care void the second they return home. Sustained cognitive stimulation and validation therapy are crucial between appointments; however, family members lack clinical training, and conventional health applications are often too convoluted for someone experiencing cognitive decline.
The consequence? Patient withdrawal, faster cognitive decline, and overwhelming caregiver fatigue.
CogniTrace bridges this divide.
CogniTrace is a voice-first cognitive support companion hosted on AWS, designed to bridge formal memory therapy with everyday home life. Leveraging Amazon Bedrock, Google Gemini, LangGraph, and RAG architectures, it transforms routine voice interactions into gentle, ongoing therapeutic reinforcement—all while providing caregivers with clear, calming visibility into daily well-being.
CogniTrace is a supportive care companion, not a diagnostic platform. It does not diagnose Alzheimer’s disease or dementia; rather, it makes the intervals between clinical appointments meaningful and supportive.



Why This Matters

Cognitive decline doesn't pause for the next clinical visit. It progresses continuously in the weeks and months in between—reflected in daily habits, forgotten words, and subtle shifts in conversation.
  • For Patients: Maintaining cognitive health through stimulation and validation therapy demands nuanced clinical techniques. Families often struggle to deliver this balance, and well-intentioned reality checks ("No, your mother passed away years ago") can trigger severe emotional distress.
  • For Caregivers: The interval between specialist visits offers virtually no structured insight into whether a loved one's condition is stable or deteriorating. Schedules and medication routines are managed via intuition and memory alone, resulting in constant psychological strain.
In regions such as India, where public awareness is limited and follow-up care is irregular, this gap is particularly pronounced. Yet even within advanced healthcare systems, continuous home-based cognitive engagement remains rare.
CogniTrace addresses these challenges by embedding cognitive support directly into daily routines. It preserves dignity and active engagement for individuals while offering structure and reassurance to caregivers. By prioritizing an audio-first interface, it meets older adults naturally where they are—eliminating the friction of navigating complex digital screens.



What We Built

CogniTrace is designed as a dual-interface ecosystem—delivering two completely segregated, role-tailored experiences on a single platform.
+-----------------------------------------------------------------------+
| COGNITRACE |
| Dual-Interface Voice-First Ecosystem |
+-----------------------------------+-----------------------------------+
| PATIENT EXPERIENCE | CAREGIVER EXPERIENCE |
| "The Gentle Companion" | "The Executive Command Center" |
+-----------------------------------+-----------------------------------+
| • Voice-guided reminiscence | • Patient summaries & schedules |
| • Personalized episodic memories | • Medication compliance tracking |
| • Soft routine orientation | • Routine-completion monitoring |
| • Naomi Feil Validation Protocol | • Longitudinal health insights |
| • Zero "test-like" anxiety | • AI-assisted care workflows |
+-----------------------------------+-----------------------------------+

1. For Patients — The Gentle Companion

An intuitive, voice-first companion designed to reinforce speech therapy and clinical memory exercises in the home:
  • Reminiscence Therapy: Voice-driven exploration using personal life stories and curated photographs.
  • Orientation & Structure: Gentle daily check-ins and soft reminders for routine tasks.
  • Non-Intrusive Engagement: Engaging cognitive exercises designed to feel natural rather than evaluative or test-like.
  • Validation Therapy Integration: Built around Naomi Feil's Validation Therapy Protocol, prioritizing emotional validation and reassuring memory recall over rigid reality orientation.

2. For Caregivers — The Executive Command Center

A comprehensive management interface offering visibility and operational support:
  • High-level patient status overviews, schedules, and clinical appointment tracking.
  • Monitoring tools for daily routine completion and medication adherence.
  • Longitudinal trend analysis powered by AI-assisted workflows.
  • Real-time insights designed to reduce caregiving stress.



How We Built This

+-----------------------------------------------------------------------------------+
| AWS CLOUD INFRASTRUCTURE |
+-----------------------------------------------------------------------------------+
| [ AWS Amplify ] Frontend (Next.js 16 / React 19 / Web Speech API) |
| │ |
| ▼ (Encrypted HTTPS via Cloudflare Tunnel) |
| [ Amazon EC2 ] Backend Orchestration (FastAPI + LangGraph Engine) |
| │ |
| ├──► [ Clinical AI Safety Guardrails ] |
| │ ├── Validation Therapy Guardrail |
| │ ├── Permanent Memory Protection |
| │ ├── Administrative Leakage Blocker |
| │ └── Schedule Mutation & Deduplication Check |
| │ |
| ├──► [ Conversational Layer ] Amazon Bedrock + Google Gemini 2.5 Flash |
| │ |
| └──► [ Context & Data Layer ] |
| ├── Amazon DynamoDB (Single-Table Partition Key Isolation) |
| └── RAG Engine (Episodic Life Stories & Photo Embeddings) |
+-----------------------------------------------------------------------------------+

AWS Cloud Architecture

CogniTrace runs natively on AWS Cloud, utilizing AWS Amplify for frontend hosting, Amazon EC2 to power the backend runtime, Amazon DynamoDB for persistent context management, and Amazon Bedrock for conversational intelligence.

1. Bifurcated Voice Engine

To preserve strict clinical and psychological boundaries, CogniTrace enforces complete separation between the Caregiver Engine and Patient Engine. System prompts, available tools, and session histories are strictly isolated. Caregiver clinical notes can never leak into patient interactions, and patients cannot unintentionally modify care schedules or personal archives.

2. Multi-Agent Orchestration via LangGraph

A LangGraph decision engine deployed on Amazon EC2 coordinates user intent resolution, safety verification, role-based tool execution, RAG retrieval, and state persistence—complete with deterministic fallbacks during external service interruptions.

3. Clinical AI Safety Guardrails

A multi-tier safety architecture evaluates requests before the LLM generates a response:
  • Validation Therapy Guardrail: Intercepts harsh reality corrections, redirecting conversations toward emotional validation and comforting personal memories.
  • Permanent Memory Protection: Prevents voice-driven deletion or alteration of core life events and photo archives.
  • Administrative Data Guardrail: Restricts clinical decline analytics and caregiver observations from being verbalized to the patient.
  • Schedule Integrity Guardrail: Prevents accidental schedule overwrites and redundant entries.

4. Conversational Intelligence & Data Persistence

  • Amazon Bedrock provides foundational reasoning, grounding, and natural conversation flow.
  • Amazon DynamoDB implements a single-table schema with strict partition-key isolation to prevent data cross-contamination, backed by an in-memory fallback for continuous availability.
  • RAG Pipeline pulls relevant family narratives and photo metadata during reminiscence exercises to anchor personal identity.

5. Engineering Challenge: Secure, Low-Latency Networking

Our HTTPS-secured frontend on AWS Amplify required communication with the Amazon EC2 backend without triggering browser-level mixed-content blocks ( ERR_MIXED_CONTENT ). Rather than setting up an Application Load Balancer with custom ACM certificates, we implemented a Cloudflare Tunnel. This created an encrypted outbound connection from EC2 to Cloudflare's edge network, enabling us to close all inbound EC2 ports, resolve mixed-content issues, and reduce connection handshake latency by approximately 35%.



What We Learned

Early clinical feedback emphasized the critical importance of clearly defining our scope. CogniTrace is intentionally non-diagnostic—this is not merely a disclaimer, but a foundational design principle. Its true clinical value lies in daily continuity: maintaining gentle, regular engagement and surfacing longitudinal patterns over time.
Our primary takeaway was recognizing the depth of the home care gap—the silent window between clinic visits where structured support vanishes. The underlying technologies (RAG, agentic LLMs, longitudinal data stores) are readily available; the missing element was an empathetic, intuitive experience families would genuinely adopt. A natural voice over complex forms. Validation over correction. That is the core purpose of CogniTrace.



Competitive Moat

CogniTrace distinguishes itself through a safety-first bifurcated architecture, clinically validated guardrails inspired by Validation Therapy, and an accessible, audio-first experience that honors patient dignity—all built on a resilient, scalable AWS infrastructure.



Business Model

  • B2B2C (Primary Channel): Neurology clinics and memory care centers provide CogniTrace to patients as a home-continuity layer, enabling objective adherence and engagement insights to inform clinical follow-ups.
  • D2C (Direct-to-Consumer): A subscription model tailored for families managing early cognitive decline at home, offering structured daily support without requiring clinical expertise.
Both models serve a unified goal: providing continuous, dignified cognitive support that enriches the time between medical appointments while safeguarding patient identity, well-being, and security.



Technology Stack

  • Frontend: Next.js 16, React 19, Tailwind CSS, Web Speech API — Deployed via AWS Amplify (CI/CD via GitHub)
  • Backend: FastAPI (Python), LangGraph Orchestration — Hosted on Amazon EC2 via Cloudflare Tunnel
  • AI & Embeddings: Amazon Bedrock, Google Gemini 2.5 Flash,  text-embedding-004 , Vector RAG Pipeline
  • Data Store: Amazon DynamoDB (Single-table architecture with partition isolation) with resilient in-memory fallback


A Commitment to Responsible AI: Ethical healthcare technology requires full transparency regarding its operational boundaries. CogniTrace does not provide medical diagnoses. Instead, it serves as the supportive infrastructure transforming voice interactions into continuous, compassionate care—preserving dignity for patients and offering peace of mind to caregivers.
Developed for CogniTrace Health. Powered by AWS Cloud (Amplify, EC2, DynamoDB, Bedrock).












References

Any opinions in this article are those of the individual author and may not reflect the opinions of AWS.
Enjoyed reading this content? Let the author know!

Your likes, comments, shares, and saves help creators reach more builders.

Loading recommendations

Loading article