Chetana..Awaken to your health

Chetana helps families track health progress. Using Amazon Nova & FHIR, we turn static reports into proactive, secure insights in your language. We don’t diagnose; we help you understand data.

Comment

Chetana (चेतना) — Project Story

🌟 Inspiration

In India, the "annual checkup" has become a ritual done for the sake of it. Millions walk away with a manila folder full of reports they cannot read, filled with medical jargon in English that doesn't translate to their daily lives. We saw a personal and pressing problem: health data is trapped in paper, and critical trends—the slow rise of blood sugar or the steady drop in hemoglobin—are missed until they become emergencies.

We built Chetana (meaning "Awareness" in Sanskrit) to turn these static observations into a living health story. Our goal isn't to replace doctors or act as a medical device; it’s to awaken people to their own health progress through the power of Amazon Nova and global FHIR standards.

🩺 What it does

Chetana is an AI health companion for the Indian family.

  • Intelligent Extraction: Simply take a photo of any lab report. Our engine extracts the data and stores it in the global FHIR R4 standard.
  • Family First: Indian health is a family affair. A single dashboard allows a user to manage records for parents, children, and themselves, with auto-matching based on patient names.
  • Proactive Insights: We don't just flag "High" or "Low." Chetana uses Extended Thinking to detect trends over time ("Your glucose has risen across three consecutive reports") and suggests follow-ups based on clinical guidelines.
  • Native Interaction: Health literacy shouldn't require English fluancy. Users can converse with their records using real-time Hindi voice or text in several Indian vernacular languages.

🏗️ How we built it

We leaned heavily into the Amazon Nova ecosystem to handle the complexity of health data:

  • Nova 2 Lite (Brain & Eyes): Acts as the multimodal engine for extracting structured data from report images and powers our Bedrock Agent for complex reasoning over health history.
  • Nova Multimodal Embeddings (Memory): Enables RAG (Retrieval-Augmented Generation) across report images and clinical guidelines, allowing the agent to "ground" its insights in real medical literature.
  • Nova 2 Sonic (Voice): Provides a bidirectional, low-latency Hindi and English voice experience, making the app accessible to the elderly or those with low digital literacy.
  • Nova Micro (Speed): Manages ultra-fast translations into Marathi, Tamil, and Bengali.
  • Architecture: Built on a serverless AWS stack using SAM, Lambda, DynamoDB (single-table design for FHIR observations), and Bedrock Guardrails to ensure PII security and SaMD safety compliance.

🚧 Challenges we ran into

  • Multimodal Nuance: Lab reports in India vary wildly in format. Training the prompt to accurately extract units (like mg/dL vs mmol/L) and map them to LOINC codes required significant iterative testing with Nova 2 Lite’s extended thinking modes.
  • Safety & Guardrails: Dealing with health data is a responsibility. We spent considerable time configuring Bedrock Guardrails to ensure the AI never provides a "diagnosis" or "treatment," but instead focuses on "interpretation" and "pointing to guidelines."
  • Hindi Polyglotism: Ensuring the voice interaction felt natural and could handle medical terminology in a mix of Hindi and English (Hinglish) was a complex fine-tuning process.

🏆 Accomplishments that we're proud of

  • Full Model Stack: We successfully integrated four different Nova models, each optimized for a specific task (Extraction, RAG, Voice, and Translation).
  • FHIR Compliance: Many hackathon projects just "talk to a PDF." We built a legitimate healthcare data pipeline that converts images into standards-compliant FHIR resources.
  • Real-Time Voice: Seeing a user ask in Hindi, "What was my last hemoglobin result?" and getting a native audio response in under 700ms felt like magic.

📚 What we learned

We learned that RAG (Retrieval-Augmented Generation) is the missing piece for health AI. By grounding the model in clinical guidelines, we drastically reduced hallucinations. We also learned that in the Indian context, "Family Management" is not a feature,it’s the foundation. Health data is rarely managed by the individual alone; it’s a shared family responsibility.

🚀 What's next for Chetana..Awaken to your health

  • Wearable Integration: Syncing lab trends with real-time heart rate and sleep data from devices.
  • Prescription Tracking: Using Nova to read handwritten prescriptions and set proactive pill reminders.
  • Community Impact: Partnering with local diagnostic labs in Tier 2/3 cities to provide "Chetana Summaries" directly to patients.
  • Expanded Languages: Bringing native voice support to more Indian languages as Sonic evolves.
Share this project:

Updates