From Predictions to Conversations: Transforming Digital Engagement for a Digital Publishing Platform

About Customer

The customer is a leading India-based digital publishing platform, delivering localized listings, classifieds, and community services. The platform enables users to discover hyperlocal content, access relevant updates, and engage with community offerings through a seamless digital experience. By combining technology, vernacular content, and data-driven insights, the organization empowers users to connect with services and information in their region efficiently and effectively.

The Digital publishing platform aimed to enhance user engagement on its astrology platform by delivering a more personalized and conversational experience. The goal was to move beyond static horoscope content and introduce an intelligent chat interface capable of providing real-time, context-aware astrological insights. The solution needed to understand user queries expressed in natural language, reference structured astrological data, and generate meaningful, human-like responses. Achieving this required designing a scalable, secure, and context-aware architecture that could integrate AI models with existing data systems, enabling smooth, intelligent, and personalized interactions across the platform.

To meet the digital publishing platform’s need for a scalable, multilingual, and cost-effective consultation model, Comprinno implemented a GenAI-powered architecture that integrates astrological intelligence with conversational AI. The solution leverages AWS Bedrock to process and interpret structured astrological data, while advanced LLMs handle natural language reasoning to deliver intelligent, context-aware responses.

A fully serverless data pipeline was built using AWS Lambda, Amazon S3, and Amazon API Gateway to manage text ingestion, rule extraction, and real-time query flow between users and the AI model. Amazon Bedrock served as the central intelligence layer, using Claude Sonnet 4 to interpret queries, reason over astrological rules, and generate accurate, contextual responses. Amazon Neptune, a graph database, was used to store structured astrological rules and relationships, enabling context-rich and accurate interpretations. DynamoDB was used for session storage, capturing conversation history and user interactions to support multi-turn, highly personalized dialogue. The system also integrated a third-party application for accurate Date-of-Birth (DOB) retrieval and used language detection services to identify the user’s preferred language, with Hinglish currently being supported for conversational responses. Guardrails were implemented to prevent the model from answering restricted or sensitive queries, ensuring safe handling of both astrological and non-astrological questions and providing graceful fallback responses. With KMS encryption, VPC isolation, and robust security controls, the solution ensured compliance and data protection. This AI-driven system transformed the astrology consultation workflow by automating routine interactions, assisting human experts with summarized insights, and delivering 24/7 personalized predictions across multiple regional languages.

Key Actions

AI Model Integration & Knowledge Processing :

• Deployed Amazon Bedrock with Claude Sonnet 4 to process Lal Kitab content, extract astrological rules, and generate structured insights.

• Implemented Bedrock Guardrails and language detection to ensure safe responses, graceful fallbacks, and Hinglish-based conversational accuracy

• Integrated Amazon Neptune (graph DB) and a third-party DOB service to deliver precise, relationship-driven astrological predictions.

AI Reasoning Engine & Knowledge Graph Orchestration

• Designed a thoughtful, multi-iteration process to convert unstructured Lal Kitab text into consistent, graph-ready entities and relationships.

• Added validation logic to ensure astrological relationships (e.g. planet-to-house mappings) remain consistent during data updates or new rule ingestion. Neptune has graph-based nodes and edges to represent planetary logic.

• Enabled dynamic confidence scoring from model outputs to help refine predictions when multiple planetary conditions overlap.

Security, Governance & Compliance :

• Enforced data encryption at rest and in transit using AWS KMS across all storage and database services.

• Applied Guardrails in Bedrock to ensure safe, relevant, and domain specific AI outputs.

• Isolated workloads within private subnets using VPC, with detailed monitoring via CloudWatch, CloudTrail, and AWS Config for governance visibility.

Conversational Experience & User Enablement :

• Developed a chat-based interface (Streamlit prototype) to test multilingual, AI-driven astrology interactions.

• Enabled multi-turn conversations and session tracking through DynamoDB for personalized user experiences.

Knowledge Enrichment & Retrieval Optimization :

• Established a clean mapping layer between user queries and the corresponding astrological rule sets in Neptune to ensure accurate retrieval.

• Added validation logic to verify that the retrieved astrological relationships align with Lal Kitab principles before they are passed to the model.

• Tuned retrieval parameters and prompts to align outputs with the domain tone and logical reasoning

Performance Monitoring & Scalability :

• Integrated Amazon CloudWatch dashboards for real-time monitoring of model latency, API performance, and query success rates.

• Adopted auto-scaling mechanisms to handle concurrent user requests without compromising response quality.

• Established a continuous improvement loop through metrics-based tuning, ensuring sustained model accuracy and optimal infrastructure costs.

Comprinno designed a secure, scalable, and serverless AWS architecture to support AI-driven data access and automation needs. The solution leverages Amazon API Gateway to handle user requests, which securely invoke AWS Lambda functions deployed across multiple availability zones for high availability and fault tolerance. These Lambda functions interact with Neptune Database for real-time data queries and updates, ensuring quick and reliable data access.

All components operate within a VPC for enhanced security, with AWS CloudWatch used for monitoring and logging system performance. AWS Systems Manager simplifies operational management, while AWS Backup ensures data protection and recovery. Security and governance are reinforced through AWS Config, AWS Security Hub, and AWS CloudTrail, providing complete visibility and compliance tracking. This architecture ensures seamless performance, automated scalability, and minimal operational overhead, aligning perfectly with the goal of delivering intelligent, on-demand data insights

Metrics for Success

The following KPIs highlight the measurable improvements achieved through the implementation of a secure, highly available, and cloud-native AWS architecture.

24/7 Availability and Consultation Capacity

• Human astrologers were available approximately 8–10 hours per day with a limited capacity of 50–60 consultations daily, leading to service bottlenecks during peak hours. Enable 24/7 consultation availability with support for 500+ concurrent consultations. The GenAI-powered astrology platform now delivers continuous 24/7 service, supporting high concurrent demand during peak usage periods without downtime or performance degradation. Measured using system uptime metrics from Amazon CloudWatch, concurrent session tracking through Amazon DynamoDB, API request volume, AWS Lambda concurrency, and performance comparisons across peak and off-peak usage periods.

Response Time and Operational Efficiency

• Users experienced 15–30 minute wait times during peak hours due to dependency on human astrologer availability. Deliver automated consultation responses within 30 seconds to provide near real-time interaction. The AI-powered consultation platform delivers average response times of 15–20 seconds for end-to-end predictions, significantly improving user experience and service responsiveness. Measured using average response latency, Lambda execution duration, Bedrock model inference time, and API request performance metrics.