Public Case Study
Building a Secure RAG-Based Knowledge
Assistant on AWS
Transforming institutional knowledge discovery and regulatory compliance for high-velocity BFSI operations.
Client Overview
Client
Industry
Use Case
Platform
Implementing Partner
Client Overview
Client
Industry
Use Case
Platform
Implementing Partner
Executive Summary
Ayan Analytics – a SEBI-registered Portfolio Management Services firm – faced mounting operational challenges in accessing SEBI circulars, master circulars, consultation papers and compliance guidance efficiently. Each manual lookup took 20 to 30 minutes, creating delays in business decisions and the risk of missing regulatory changes
Infimatrix designed and deployed a Retrieval-Augmented Generation (RAG) Knowledge Assistant using Amazon Bedrock (Nova Pro) and Amazon Bedrock Knowledge Bases on a fully serverless AWS architecture. Employees now ask questions in natural language and receive source-cited answers in 2.6 seconds on average, with ~90% answer accuracy on a benchmark of regulatory questions. Every response is checked by Amazon Bedrock Guardrails and secured through Amazon Cognito authentication, AWS IAM least-privilege policies and AWS KMS encryption.
The Challenge
Manual Search Inefficiency
Employees spent 20–30 minutes per query searching shared drives, email threads and physical files, often consulting 2–3 colleagues.
Knowledge Silos
Critical institutional knowledge was locked in documents accessible only to specific teams.
Expert Dependency
Junior staff escalated 8–12 queries per week to a few senior experts, creating bottlenecks and single points of failure.
Information Overload
Documents were scattered across shared drives, email threads and physical files, and finding the right answer required knowing which document to look in.
Compliance Risk
Updated SEBI circulars could sit unread for weeks, increasing non-compliance exposure.
The Solution
Infimatrix implemented a secure, fully serverless RAG architecture on AWS in the Mumbai Region (ap-south-1), defined entirely in AWS CloudFormation. Users authenticate through Amazon Cognito, and the platform is monitored through Amazon CloudWatch (12 custom metrics, 8 alarms) and audited through AWS CloudTrail.
Documents uploaded to Amazon S3 trigger an Amazon EventBridge rule that starts an AWS Step Functions workflow. The workflow validates each file, extracts text from scanned PDFs with Amazon Textract, classifies the document by department with a custom Amazon SageMaker model, enriches its metadata with Amazon Comprehend, records it in an Amazon DynamoDB document registry, and syncs it into Amazon Bedrock Knowledge Bases, which chunks, embeds (Titan Embeddings V2) and indexes it in Amazon OpenSearch Serverless for hybrid (semantic + keyword) search. Amazon SNS sends a notification when each new document is indexed. When an employee asks a question, an AWS Lambda orchestrator retrieves the most relevant passages and invokes Amazon Bedrock (Nova Pro) to generate a source-cited answer, which is checked by Amazon Bedrock Guardrails before delivery. Every query is logged in DynamoDB for audit.
Responsible AI — Amazon Bedrock Guardrails
In regulated industries, an inaccurate compliance response is a liability. Infimatrix configured Amazon Bedrock Guardrails to help ensure AI responses are safe, grounded, and within business scope – significantly reducing the risk of ungrounded or inaccurate responses through grounding checks and source attribution.
Guardrail
Configuration
Purpose
Content Filters
Block hate, insults, sexual, violence and misconduct content
Responsible AI in regulated environments
Denied Topics
Block requests for investment advice (buy/sell/hold recommendations, price predictions) and competitor comparisons
Avoids regulatory liability for a PMS firm
PII Masking
Mask names, emails, phone numbers, bank account numbers, PAN and Aadhaar; block card and SSN numbers
Data privacy for sensitive financial data
Contextual Grounding
Grounding threshold 0.5 + relevance threshold 0.5
Withholds answers not supported by source documents
Prompt Attack Detection
High strength on user input
Prevents prompt injection and jailbreak attempts
Profanity Filter
Managed profanity word list on input and output
Maintains professional communication standards
Business Outcomes
KPI
Baseline
Results
Time to find information
20-30 minutes
2.6 seconds average (P90 4.8 s)
Answer accuracy
Not measurable
~90%
User satisfaction
Not tracked
96.8% positive
Document coverage
No centralised access
50 regulatory documents searchable
Knowledge base freshness
Weeks
Median 15 seconds, always under 5 minutes
Senior staff escalations
8–12 per week
2–3 per week (65–70% reduction)
Key Benefits
- Strengthened Compliance Posture: Every answer is traceable to a source document and every query is logged for audit, supporting BFSI audit requirements.
- Eliminated Knowledge Silos : Organisation-wide, role-based access to SEBI circulars and compliance guidance – regardless of owning team.
- Reduced SME Dependency : Senior staff escalations fell 65–70%, freeing experts for higher-value work..
- Accelerated Onboarding : New employees self-serve answers instead of waiting for training sessions or colleagues.
- Scalable & Future-Ready : Fully serverless and defined in AWS CloudFormation; new documents are auto-indexed within minutes with no manual effort.
- Strengthened Compliance Posture : Every answer is traceable to a source document and every query is logged for audit, supporting BFSI audit requirements.
- Eliminated Knowledge Silos : Organisation-wide, role-based access to SEBI circulars and compliance guidance – regardless of owning team.
- Reduced SME Dependency : Senior staff escalations fell 65–70%, freeing experts for higher-value work..
- Accelerated Onboarding : New employees self-serve answers instead of waiting for training sessions or colleagues.
- Scalable & Future-Ready : Fully serverless and defined in AWS CloudFormation; new documents are auto-indexed within minutes with no manual effort.
Conclusion
Infimatrix successfully deployed a secure, enterprise-grade RAG knowledge assistant for Ayan Analytics on AWS, built on three core pillars:
Security-First Design
Serverless-native security posture with no VPC or EC2 attack surface, authenticated access via Amazon Cognito and least-privilege AWS IAM roles scoped per Lambda function, encryption at rest via AWS KMS and in transit via TLS, AWS WAF on CloudFront with managed rules and rate limiting and a full audit trail via AWS CloudTrail.
Responsible AI
Amazon Bedrock Guardrails ground every response in approved documents, significantly reducing the risk of inaccurate or ungrounded answers - critical for BFSI compliance.
Measured Business Outcomes
20–30 minute manual searches replaced by 2.6-second AI responses, with ~90% answer accuracy, 96.8% positive user feedback and a 65–70% reduction in SME escalations.
This fully managed, serverless solution is applicable to any BFSI organisation seeking to modernise knowledge management, strengthen compliance posture, and break down knowledge silos.
Conclusion
Infimatrix successfully deployed a secure, enterprise-grade RAG knowledge assistant for Ayan Analytics on AWS, built on three core pillars:
Security-First Design
Serverless-native security posture with no VPC or EC2 attack surface, authenticated access via Amazon Cognito and least-privilege AWS IAM roles scoped per Lambda function, encryption at rest via AWS KMS and in transit via TLS, AWS WAF on CloudFront with managed rules and rate limiting and a full audit trail via AWS CloudTrail.
Responsible AI
Amazon Bedrock Guardrails ground every response in approved documents, significantly reducing the risk of inaccurate or ungrounded answers - critical for BFSI compliance.
Target Business Outcomes
20–30 minute manual searches replaced by 2.6-second AI responses, with ~90% answer accuracy, 96.8% positive user feedback and a 65–70% reduction in SME escalations.
This fully managed, serverless solution is applicable to any BFSI organisation seeking to modernise knowledge management, strengthen compliance posture, and break down knowledge silos.
About Infimatrix
Infimatrix Technologies Pvt. Ltd. is an AWS Partner specialising in Generative AI, cloud-native architecture, and digital transformation for regulated industries.
Specialisations:
Generative AI
RAG Solutions
Amazon Bedrock
BFSI Digital Transformation
About Infimatrix
Infimatrix Technologies Pvt. Ltd. is an AWS Partner specialising in Generative AI, cloud-native architecture, and digital transformation for regulated industries.