Amazon Bedrock AI Chatbot for Fund Administration | Case Study | Logicata

Executive Summary

A US fund administration platform managing over $12 billion in assets had built an AI chatbot to let fund managers query their data in natural language, but the application had nowhere to land: Amazon Bedrock wasn’t enabled, no IAM scoped for it, no VPC endpoint, no observability. Logicata delivered the missing AWS infrastructure via Terraform, consistent with the customer’s existing IaC practices. The chatbot launched into production with all AI traffic staying inside the VPC and full visibility into model usage and cost.

Customer Overview

Sector: Financial Services

Location: United States

The customer is a US-based private market fund administration company. Their platform manages over $12 billion in assets under administration for more than 200 fund managers, providing fund accounting, investor relations, and operational tooling. Fund managers spend significant time manually searching documents and reports. An AI chatbot promised to collapse that time, but the application couldn’t talk to Bedrock without the right AWS scaffolding in place.

The Challenge

The development team had built the chatbot application logic. The AWS infrastructure to connect it to Bedrock was a different problem:

• No model access. Claude Haiku and Sonnet were not enabled in the target account and region.
• No private network path. AI invocations against sensitive financial data could not be allowed to traverse the public internet.
• Existing Terraform codebase. Anything new had to match the customer’s IaC practices, not introduce a parallel snowflake.
• No observability. Zero visibility into model usage, latency, error rates, or accumulating spend.
• IAM done right or not at all. The ECS task role running the chatbot needed precisely scoped bedrock:InvokeModel permissions, not a sweeping policy that the next security review would flag.

The Solution

Logicata delivered targeted Terraform-native AWS infrastructure to unblock the chatbot:

1. Bedrock model access. Enabled Claude Haiku and Sonnet in the account and region the chatbot needed.
2. Scoped IAM. Added bedrock:InvokeModel with resource-specific permissions to the existing ECS task role. No new roles, no new users, no broader policy than required.
3. VPC interface endpoint for Amazon Bedrock. All chatbot AI traffic now stays inside the private network. Sensitive financial data never touches the public internet.
4. CloudWatch dashboards. Real-time visibility into invocations, latency, error rates, and token consumption.
5. CloudWatch alarms. Configured for throttling, error rate thresholds, and cost anomaly detection so unexpected chatbot spend gets flagged rather than discovered on the bill.

Every change was delivered as Terraform, fitting the customer’s existing codebase. The customer owns and can evolve the infrastructure without further engagement.

Results

• Chatbot in production. Fund managers can now query financial data using natural language.
• All AI traffic stays inside the VPC. Sensitive financial data never leaves the private network for model invocation.
• Full observability. CloudWatch dashboards cover usage, latency, errors, and token consumption in real time.
• Cost surprises eliminated. Anomaly detection alarms fire before runaway usage shows up on the invoice.
• Customer-owned infrastructure. Terraform-native delivery means the customer can manage and extend the chatbot platform independently.

AWS Services Used

• Amazon Bedrock (Claude Haiku, Claude Sonnet)
• Amazon VPC (Interface Endpoints)
• AWS IAM
• Amazon CloudWatch (Dashboards, Alarms)
• Amazon ECS (existing task role)
• Terraform

About Logicata

Logicata is an AWS Advanced Partner holding the AWS Cloud Operations Management Competency, validated through an independent third-party audit. Logicata helps organisations build and operate secure, well-governed cloud platforms on AWS, enabling customers to reduce operational risk, meet assurance expectations, and scale with confidence.