Every single day across India, millions of patients walk out of outpatient clinics overwhelmed. A typical clinic visit in India lasts less than ten minutes. Patients leave with handwritten prescriptions, complex dosage schedules, dietary warnings, and follow-up dates. Within hours, more than half of those critical instructions are forgotten or misunderstood.
Standard AI chatbots cannot solve this problem safely because healthcare demands absolute accuracy. If an AI hallucinates a drug dosage, it directly puts patient lives at risk.
To solve this healthcare memory gap, I built YAAD. The name comes from the Hindi word Yaad, meaning to remember. It is an evidence-first post-visit assistant designed specifically for patients and families in India.
The entire system runs on an AWS serverless architecture deployed in Stockholm:
First, consultation audio is uploaded to Amazon S3, protected by customer-managed AWS KMS encryption with thirty-day lifecycle rules.
Second, Amazon Transcribe runs speech-to-text with the en-IN language model and two-speaker diarization, cleanly separating what the doctor said from what the patient asked.
Third, AWS Step Functions coordinates the entire pipeline asynchronously, eliminating the need for idle servers.
Fourth, Amazon Bedrock running Nova Lite extracts clinical summaries and powers the Ask YAAD conversational drawer. Output is capped at 256 tokens to ensure fast, concise answers and complete cost predictability.
Fifth, Amazon DynamoDB stores visit metadata, transcripts, and evidence mappings with on-demand scaling.
Finally, AWS Lambda and Amazon API Gateway deliver the web application and HTTPS REST API in a single serverless deployment.
YAAD turns fast-paced clinical visits into calm, organized care plans that patients and caregivers can actually rely on.
Loading article