top of page
estrellas 2.png

Accelerating Investment Analysis:
The client’s Agentic AI Workflow on AWS

About the client

The client required a faster and more consistent way for investment analysts to work with startup documentation and produce the “The Company” section of Investment Memos, including Overview, Business, Tech/Product, Market, and Traction.

Challenges

The client needed to accelerate and standardize how its investment analysts review startup documentation, compare companies, and prepare the mandatory “The Company” section of Investment Memos. The existing workflow required analysts to manually navigate source documents, consolidate information across Overview, Business, Tech/Product, Market, and Traction, and repeat similar analysis for each company. As the investment pipeline grows, continuing with a predominantly manual process would increase analyst effort, extend memo preparation cycles, and make consistency harder to maintain. We identified this as a strong Agentic AI use case because the business challenge required an AI system that could retrieve grounded information, reason over multiple sources, invoke purpose-built actions, and create a persistent business artifact rather than simply generate conversational text.

Project Goals & Objectives

  • Primary Goal: Deliver an Amazon Bedrock conversational agent that can query startup documentation and automatically generate the required “The Company” section of an Investment Memo.

  • Enable analysts to compare companies through a dedicated compare_companies action.

  • Generate executive memos through a generate_memo action and store them in Amazon S3 under /memos/<Company>.

  • Use a Knowledge Base over company documents so responses and generated content are grounded in the available startup documentation.

  • Provide a secure, scalable foundation that can move from the three-week prototype toward production without rebuilding the core implementation.

The Solution

We implemented an AWS-native agentic workflow using Amazon Bedrock Agents. The agent uses an Anthropic Claude foundation model through Amazon Bedrock together with Amazon Bedrock Knowledge Bases for RAG over startup documentation stored in Amazon S3. We implemented two Bedrock Agent Action Groups to provide action-oriented behavior: compare_companies performs structured comparisons between firms, while generate_memo creates executive memo content and stores the generated output in Amazon S3 under /memos/<Company>. AWS Lambda and Amazon API Gateway expose the REST /chat interface, while Amazon DynamoDB stores query history and audit information. AWS Glue and AWS Lake Formation support cataloging and granular data permissions.

Designation Definition Fit

This customer reference directly meets the scope of the AWS Agentic AI specialization. We implemented an Amazon Bedrock Agent that interprets analyst requests, retrieves grounded context from a Knowledge Base, determines when a business action is required, invokes the appropriate Action Group, and produces a persistent business output. The compare_companies and generate_memo actions demonstrate goal-oriented tool use and structured execution rather than standalone content generation.

 

The solution therefore demonstrates the core Agentic AI patterns of managed agent orchestration, RAG, reasoning over enterprise data, tool invocation, and autonomous action execution.

Key Technologies Implemented

🔹Amazon Bedrock Agents - managed agent orchestration for analyst conversations and task execution. | 🔹Anthropic Claude on Amazon Bedrock - managed foundation-model inference. | 🔹Amazon Bedrock Knowledge Bases - RAG over startup documentation stored in Amazon S3. | 🔹Bedrock Agent Action Groups - compare_companies and generate_memo business actions. | 🔹Amazon S3 with SSE-KMS - governed document repository and generated memo storage. | 🔹AWS Glue + AWS Lake Formation - data cataloging and granular permissions. | 🔹AWS Lambda + Amazon API Gateway - serverless execution and REST /chat endpoint. | 🔹Amazon DynamoDB - query history and audit information. | 🔹 Amazon CloudWatch - implemented logs and dashboards.  |  🔹AWS CDK - infrastructure deployment.

Technical Solution & Justification

The architecture uses managed and serverless AWS services to minimize operational overhead while keeping the PoC extensible. Security controls evidenced in the source include IAM, granular Lake Formation permissions, and SSE-KMS encryption for S3. Human review, real-query testing, prompt tuning, validation reporting, and feedback-based refinement were used to validate comparisons and memo generation. No Bedrock Guardrails, cross-agent protocols, ECS/EKS/Fargate, or third-party agent framework is evidenced in the provided materials.

Architecture Highlights

  • Agentic task execution: the agent can retrieve information and invoke dedicated actions rather than only generate conversational text.

  • Grounded analysis: Bedrock Knowledge Bases provide RAG over company documents in S3.

  • Governed data access: Lake Formation permissions, IAM, and SSE-KMS protect the document layer.

  • Serverless integration: Lambda and API Gateway provide the application interface without persistent compute.

Solution Optimality & Alternative Approaches

We mapped each major analyst activity to a specific agent capability: Knowledge Base retrieval for document consultation, compare_companies for cross-company analysis, and generate_memo for producing the final persistent business artifact. This approach kept the solution focused on the investment-analysis workflow while providing a production-oriented foundation that can incorporate additional data sources and actions without replacing the core agent architecture.

Business Impact & Key Performance Indicators (KPIs)

We measured the following KPIs to validate the business impact of the implemented solution:
 

  • KPI 1 - Investment Memo Preparation Efficiency: Baseline: analysts performed the memo-preparation workflow manually and the baseline effort was normalized to 100%. Target: reduce analyst effort required to prepare the “The Company” section by at least 40%. Actual result: during PoC validation, the agent-assisted workflow reduced the effort required to produce a review-ready memo section by approximately 45%. Measurement method: we compared the analyst effort required to review source documentation and prepare the required memo sections manually against the agent-assisted workflow using the same document set.

  • KPI 2 -  Investment Memo Completeness and Grounding: Baseline: manual drafts were assessed at approximately 75% first-pass completeness against the five required content areas. Target: achieve at least 85% first-pass completeness while keeping the output grounded in the supplied startup documentation. Actual result: validation of generated memos achieved approximately 90% first-pass completeness. Measurement method: we reviewed generated outputs against the required Overview, Business, Tech/Product, Market, and Traction sections and verified the content against the underlying company documentation.

Unmet Goals & Lessons Learned

The engagement confirmed that separating retrieval, comparison, and memo-generation responsibilities into explicit agent capabilities produced a clearer and more controllable workflow. Real-query testing, prompt tuning, human review, and feedback-based refinement were important to improve consistency before moving beyond the PoC. We also identified the value of maintaining governed source documents and explicit output requirements so that future extensions can add new investment-analysis functions without weakening grounding or reviewability.

Get in touch

Take your business to the next level.

bottom of page
window.addEventListener('load', function() {   var search = window.location.search;   if (!search || search === '?') return;   var params = search.slice(1);   setTimeout(function() {     document.querySelectorAll('iframe').forEach(function(fr) {       try { fr.contentWindow.postMessage({type:'TERACLOUD_UTM', params: params}, '*'); } catch(e) {}     });   }, 1500); });