Case Description Client is a AI native spiritual services company, needed a full-stack platform to book Hindu rituals and venerations at home — AI-powered call transcription, multi-role admin, Priest onboarding, geofenced check-in/out, payments etc. Learn more ! Get on a 1:1 call with our experts to discuss how Generative AI can add value to your o
AI-native FaithTech Mobile App on AWS Case Description: The client is an AI-native spiritual services company that required a full-stack platform for booking Hindu rituals and home venerations. Core needs included AI-powered call transcription, multi-role administration, Priest onboarding, geofenced check-in/out, and payment processing. Problem Statement: The client aimed to bring Vedic puja services entirely online — covering browsing, booking, consulting an Acharya by phone, samagri (ritual materials) logistics, in-home service delivery, payments, and post-service ratings. Operational requirements included: - Multi-step, multi-party coordination: customer, Acharya (priest), booking executive, content creator, logistics executive, customer success, and super-admin each needed purpose-built views and task queues. - Voice-first pre-booking: Indian customers expect to speak with a priest to confirm ritual details (date, time, deity, occasion, address) before paying; that call serves as the source of truth for booking details. - Language complexity: customer-Acharya conversations occur in Hindi, frequently mixed with Sanskrit ritual terminology; booking executives then extract structured fields from those conversations to create the booking. - Compliance and trust: payment flows, refunds, cancellation windows, geo-locked service delivery, and traceable records are all required to build customer trust. - Operational scale-out: as Acharya, samagri, and customer volumes grew, the platform needed role-based access control, task automation, and notification orchestration. Proposed Solution & Architecture (AWS ap-south-1 Mumbai Region): Compute Tier: - Amazon ECS on AWS Fargate: containerized Node.js backend (app-api, 2 tasks min / 10 max, 2 vCPU / 4 GB per task) behind an Application Load Balancer with HTTPS/ACM certificates; WAF attached using AWS Managed Rules (Core Rule Set, SQL injection, XSS). - AWS Lambda for: Exotel webhook callback handler (call-status updates to booking state transitions); S3 to Transcribe async trigger on call-recording upload; Razorpay webhook handler (payment status reconciliation); Scheduled reminders via Amazon EventBridge cron (8 AM IST daily). - Amazon API Gateway (REST API) fronting the Node.js backend with API keys for partner integrations; throttling set at 1,000 requests/second/API key. Data Tier: - Amazon RDS for PostgreSQL: Multi-AZ, db.m6g.large, 100 GB gp3 storage auto-scaling to 1 TB, automated daily backups with 14-day retention, Performance Insights enabled. - Amazon ElastiCache for Redis: cache.t4g.small, cluster mode disabled with replica; used for puja-listing cache, user-session store, and rate limiting. - Amazon S3 buckets (all SSE-KMS with customer-managed CMK, Object Lock governance mode on invoices for 7-year retention per Indian GST audit requirements): app-recordings (Exotel call audio, lifecycle to S3 Glacier Instant Retrieval after 90 days), app-checklists (puja checklist PDFs, public via CloudFront signed URLs), app-invoices (Razorpay-linked invoice PDFs), app-samagri (product imagery), app-selfies (Acharya check-in selfies, lifecycle to Glacier Deep Archive after 180 days). - Amazon CloudFront: distribution with OAC serving images and PDFs from S3; signed URLs for private content. AI / Voice / Notifications Tier: - Amazon Transcribe: Hindi (hi-IN) transcription of Exotel call recordings; custom vocabulary trained on Sanskrit ritual terminology, Acharya names, and puja types; async StartTranscriptionJob triggered on S3 PutObject; typical transcription latency approximately 1 minute. - Amazon SNS: topic per lifecycle event (booking-created, payment-confirmed, acharya-allocated, samagri-dispatched, cancellation, refund-processed); fan-out to SQS subscribers for downstream workers. - Exotel Calling API (third-party CPaaS): Leg1 to Leg2 call flow with status webhooks POSTed to API Gateway then Lambda. - Razorpay (third-party payment gateway): webhook-driven payment reconciliation; idempotency via DynamoDB. - WhatsApp Business API via third-party BSP (Gupshup/AiSensy): templated messages in Hindi/English for Welcome, Booking Confirmation, Payment Confirm, Invoice, Acharya Allocate (Hindi to priest), Samagri Dispatch, Cancellation, Refund. - Firebase Cloud Messaging (FCM): for iOS/Android push notifications; messages published via SNS to Lambda to FCM HTTP v1 API. Identity & RBAC: - Amazon Cognito User Pools: separate pool for User App (OTPless + Google federation) and Admin Panel (MFA required for super-admin). - OTPless third-party integration for Indian-number-friendly OTP and WhatsApp auth flows. - Admin Panel RBAC enforced via Cognito Groups to IAM roles; row-level access in Postgres keyed to user_id + role. Observability: - Amazon CloudWatch Logs. - Amazon CloudWatch RUM (Real User Monitoring) on React admin panel for frontend error tracking. - AWS X-Ray: traces across ECS to Lambda to RDS. Networking & Security: - Amazon VPC: public subnets (ALB) and private subnets (ECS tasks, RDS, ElastiCache); NAT Gateway for outbound; VPC Endpoints for S3, DynamoDB, and Secrets Manager. - AWS WAF: on CloudFront and ALB; rate-based rules block more than 100 requests per 5 minutes from a single IP. - AWS KMS: customer-managed CMKs; AWS Secrets Manager for Razorpay keys, Exotel keys, WhatsApp BSP tokens, and FCM server keys (all auto-rotated). - AWS CloudTrail: for audit; AWS Config rules; Amazon GuardDuty enabled for threat detection. Project Outcomes and Success Metrics: a21.ai delivered a full-stack platform spanning three surfaces: - User App (iOS & Android) - Acharya App (iOS & Android) - Admin Panel (web app) - A consumer-facing web app All surfaces backed by a unified AWS backend. Additional deliverables: Versioned releases (Release Notes v1.0.3, Progress Report v1.0.4), active testing regimen (Mobile user testing notes), marketing funnel instrumentation for conversion tracking, GA KPI framework, operational run assets including error logs for user app and promotion module workflow video (v1.0.3). 3-Year TCO Analysis & Savings: For a Vedic-puja marketplace in India, no existing SaaS stack covers Hindi-vernacular support, Acharya geofencing, and booking-lead extraction. The custom AWS-native build is the only viable architecture; ROI is driven by retention uplift from trust-building features — check-in, call recordings, and transparent pricing — in a trust-dependent category. Total 3-year TCO: approximately $310,000 (including build and maintenance costs): - Build: $50,000 - Maintenance: $20,000/year ($60,000 over 3 years) - AWS Running Costs: $200,000 (growing from $800/month in Year 1 to $12,000/month in Year 3)