
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
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 powerful cognitive therapy wasn't confined to a clinic — but woven into an ordinary conversation at the kitchen table?
More than 57 million people were living with dementia in 2021, with nearly 10 million new cases each year — and more than 60% of them in low- and middle-income countries ( WHO ). Dementia is most common after 65, but young-onset dementia accounts for up to 9% of cases.
Here is the part no one talks about: individuals in early-stage Alzheimer's and Mild Cognitive Impairment (MCI) diligently attend memory clinics and speech-therapy sessions — and then a silent care gap opens the moment they walk back through their front door. Cognitive exercise and validation therapy are vital between appointments, but family members aren't trained clinicians, and mainstream health apps are far too complex for someone experiencing cognitive decline.
The result? Disengagement. Accelerated cognitive stagnation. And caregivers quietly burning out.
CogniTrace is here to change that.
CogniTrace is a voice-first cognitive-care companion deployed on AWS that bridges clinical memory therapy and daily home life. Powered by Amazon Bedrock, Google Gemini, LangGraph, and RAG, it turns everyday voice interaction into gentle, continuous therapeutic support — while giving caregivers the visibility they need to feel calm instead of overwhelmed.
CogniTrace is a care-support platform, not a diagnostic system. It does not diagnose dementia or Alzheimer's disease. It makes the time between clinical visits meaningful.
Why This Matters
Cognitive decline doesn't wait for the next appointment. It happens in the weeks and months in between — in daily routines, in the words that go missing, in the conversations that slowly change.
- For patients, consistent cognitive stimulation and validation therapy require specialized clinical technique. Families want to help, but they aren't trained to deliver it — and a "wrong" response ("No, your mother passed away years ago") can cause real distress.
- For caregivers, the gap between clinic visits offers no structured way to know whether their loved one is stable or slipping. Medication and routine decisions get made on memory and intuition rather than data — and the emotional load is relentless.
- In regions like India where awareness is low and follow-up is inconsistent, this gap is even wider. But even in advanced healthcare systems, continuous home-based cognitive support is still uncommon.
CogniTrace addresses this by bringing cognitive care into everyday life. For individuals, it enables dignity and daily engagement. For caregivers, it provides reassurance and structure. Accessibility is central: a voice-first design meets elderly users where they are, without asking them to learn a complicated app.
What We Built




CogniTrace is a dual-interface ecosystem — one platform, two completely isolated experiences.
1. For Patients — The Gentle Companion
An accessible, audio-first therapeutic companion that reinforces memory-clinic and speech-therapy work at home:
- Voice-guided reminiscence using personalized episodic memories and photos
- Gentle routine orientation and daily reminders
- Non-intrusive cognitive stimulation that never makes the patient feel "tested"
- Built on Naomi Feil's Validation Therapy Protocol — it validates emotions first, and anchors identity through comforting memory retrieval instead of harsh reality correction
2. For Caregivers — The Executive Command Center
- Patient summaries, care schedules, and appointment management
- Medication compliance and routine-completion tracking
- Longitudinal insights and AI-assisted workflows
- Actionable peace of mind
How We Built This

AWS Architecture
CogniTrace runs entirely on AWS Cloud, using AWS Amplify for the web experience, Amazon EC2 for the backend engine, Amazon DynamoDB for persistent context, and Amazon Bedrock for the conversational AI layer.
┌───────────────────────────────────────────────────────────────────────────────┐
│ COGNITRACE AWS ARCHITECTURE │
├───────────────────────┬────────────────────────┬──────────────────────────────┤
│ AWS AMPLIFY │ AMAZON EC2 │ AMAZON DYNAMODB │
│ • Next.js Frontend │ • FastAPI Backend │ • User Profiles │
│ • GitHub CI/CD │ • LangGraph Agent │ • Chat History │
│ • HTTPS / CDN │ • Voice Processing │ • Reminders & Appointments │
│ │ • Speech Analysis │ • Memories / Vector Context │
└───────────┬───────────┴────────────┬───────────┴──────────────┬───────────────┘
│ HTTPS │ │
▼ │ ▼
┌───────────────────────┐ │ ┌──────────────────────┐
│ CLOUDFLARE TUNNEL │────────────┘ │ RAG / CONTEXT │
│ Secure HTTPS Bridge │ │ Patient-specific │
└───────────────────────┘ │ Retrieval │
└──────────┬───────────┘
▼
┌──────────────────────────────┐
│ AI AGENT: Amazon Bedrock + │
│ Gemini + LangGraph + │
│ Clinical Guardrails │
└──────────────────────────────┘1. Bifurcated Voice Engine
To guarantee clinical and psychological safety, CogniTrace enforces strict separation between the Caregiver Engine and the Patient Engine — across system prompts, permitted tools, and database session histories. A caregiver's clinical notes can never leak into a patient's conversation, and a patient can never accidentally delete a reminder or a sacred memory.
Dimension | Caregiver Agent | Patient Agent |
|---|---|---|
Persona | Professional clinical coordinator | Gentle validation companion |
Session Isolation | CAREGIVER_CHAT# (DynamoDB) | PATIENT_CHAT# (DynamoDB) |
Permitted Actions | Track compliance, manage reminders & appointments | View reminders, mark meds done, hear progress, view memories |
Restricted | Cannot delete patient memories | Cannot delete/edit schedules or memories |
2. LangGraph Multi-Agent Orchestration
A LangGraph decision engine on Amazon EC2 handles intent resolution, clinical guardrail checks, role-scoped routing, RAG memory retrieval, and persistence — with a deterministic fallback if any service is offline.
3. Clinical AI Guardrails
A multi-layered safety net running before the LLM ever responds:
- Validation Therapy Guardrail — blocks harsh reality correction, responds with emotional validation + photo memories
- Permanent Memory Protection — life stories and photos cannot be deleted by voice
- Administrative Leakage Guardrail — clinical decline metrics never spoken to patients
- Schedule Mutation & Deduplication Guardrails — prevents accidental edits and duplicate entries
4. Conversational Layer & Data
- Amazon Bedrock powers the conversational reasoning and grounding layer
- Amazon DynamoDB uses a single-table design with partition-key isolation for zero cross-context leaks, plus a resilient in-memory fallback for zero downtime
- RAG retrieves top matching family life stories and photos to anchor identity during reminiscence
5. Solving a Real Deployment Challenge
Our AWS Amplify frontend (HTTPS) couldn't talk directly to our Amazon EC2 backend (raw HTTP) — browsers blocked it with
ERR_MIXED_CONTENT . Rather than provision an expensive Load Balancer + ACM certificates, we used a Cloudflare Tunnel: an encrypted outbound bridge from EC2 to Cloudflare's edge. This closed all inbound ports on EC2, eliminated mixed-content errors, and reduced handshake latency by ~35%.
What We Learned
The feedback that resonated most was about clinical scope — and it forced us to be precise about what we were actually claiming. CogniTrace is not a diagnostic tool. That is not a disclaimer; it is the product. The value isn't in a single reading — it's in continuity, in showing up gently every day, in the trend that emerges over weeks.
The deepest insight wasn't about the AI. It was about the gap — the silent stretch between clinic visits where cognitive care simply stops. The infrastructure to close that gap (conversational AI, RAG, longitudinal context) already exists. What was missing was something families would actually use. A voice, not a form. Validation, not correction. That's what CogniTrace tries to be.
Competitive Landscape
Product | What it does | CogniTrace's Difference |
|---|---|---|
Generic voice assistants (Alexa/Google) | General-purpose tasks & reminders | No Validation Therapy — can deliver harmful reality correction |
Standard dementia care apps | Task lists, medication logging | Complex UI unsuited to cognitive decline; not voice-first |
Reminiscence therapy apps | Static photo albums | No conversational AI, no RAG-driven personalization |
Clinical cognitive assessment tools | Enterprise/clinical screening | Clinician-only; no continuous home companion or caregiver bridge |
CogniTrace's moat: a safety-first bifurcated architecture, clinically-grounded Validation Therapy guardrails, and a voice-first design built for dignity — all on scalable AWS infrastructure.
Business Model
The primary channel is B2B2C — memory clinics and neurology practices offer CogniTrace to patients as a continuity-of-care layer between visits, with structured engagement and compliance data flowing back into clinical review.
A direct-to-consumer tier targets families already managing a diagnosis at home who need daily support and reassurance without clinical training.
Both channels feed the same value: continuous, dignified cognitive care that makes the time between clinical visits medically meaningful — while keeping patient memories, dignity, and safety at the absolute center.
Technology Stack
- Frontend: Next.js 16, React 19, Tailwind CSS, Web Speech API — deployed on AWS Amplify (CI/CD from GitHub)
- Backend: FastAPI (Python), LangGraph orchestration — deployed on Amazon EC2 + Cloudflare Tunnel
- AI: Amazon Bedrock + Google Gemini 2.5 Flash,
text-embedding-004, RAG / vector retrieval
- Data: Amazon DynamoDB (single-table, partition isolation) with resilient in-memory fallback
Responsible healthcare technology must be transparent about what it cannot do. CogniTrace does not diagnose. It is the infrastructure that turns everyday voice into continuous, compassionate cognitive care — for the patients who deserve dignity, and the caregivers who deserve peace of mind.
Built for CogniTrace Health. Powered by AWS Cloud — Amplify, EC2, DynamoDB, and Bedrock.
Enjoyed reading this content? Let the author know!
Your likes, comments, shares, and saves help creators reach more builders.
Loading recommendations
Loading article