AutoRFP AgentCore

An AI-Powered End-to-End RFP Automation System on AWS

Comment

Inspiration

In large presales and proposal teams, responding to complex RFPs is a slow, manual, and error-prone process. It involves reading hundreds of pages, clarifying ambiguous requirements, estimating costs, and drafting SOWs — often taking days. We wanted to leverage AWS Bedrock’s reasoning LLMs and AgentCore to create an autonomous solution that saves time, improves accuracy, and delivers explainable decisions On the sales side, our solution standardizes SOW creation, accelerates pricing and ARR calculations, and leverages AI to focus team efforts on high-value activities, increasing win rates and reducing time-to-proposal. This fusion of smart integration and automated decision-making reimagines retail efficiency and sales effectiveness for the modern enterprise.

What it does

AutoRFP AgentCore is a fully autonomous, AI-driven multi-agent system that automates the end-to-end presales workflow:

  1. RFP Parsing Agent – Reads and structures complex PDF RFPs using Claude 3.5 Sonnet.
  2. Clarification Agent – Generates precise questions to eliminate ambiguity.
  3. Pricing & Funding Agent – Estimates cost ranges and feasibility using reasoning LLMs.
  4. SOW Drafting Agent – Creates professional Statements of Work in DOCX format.
  5. Architecture Agent - Creates Architecture Diagrams for the flow of work in .png format. Orchestrator Agent – Coordinates all agents through AWS Lambda + SQS, ensuring traceability and scalability. The result: a 65 % faster RFP turnaround time with fully explainable AI outputs stored securely in Amazon S3.

How we built it

  1. AWS Bedrock + Claude 3.5 Sonnet for reasoning and structured text generation.
  2. AgentCore Runtime to orchestrate agents, manage flows, and enable observability.
  3. S3 + Lambda + AgentCore + Cognito for event-driven automation and asynchronous agent communication.
  4. Python / SageMaker for backend development and testing.
  5. React UI (to be deployed) for intuitive document uploads and real-time status tracking.

Challenges we ran into

  1. Working in a limited accesible IAM roles in our VPC network.
  2. Working with strands Agent for the first time.
  3. Coding backend entirely in Sagemaker notebook and limited access to AWS Services.

Accomplishments that we're proud of

  1. Packaging and deploying multi-agent logic within Lambda size limits.
  2. Leveraged MCP-Servers while developing the AWS Architecture Strands agent.
  3. Ensuring generated SOWs remain domain-accurate and audit-ready.

What we learned

  1. How to design agentic architectures using AWS Bedrock + AgentCore.
  2. How to leverage inference profiles, and LLM reasoning for autonomous decision pipelines.
  3. The importance of explainability and traceability in enterprise AI adoption.

What's next for AutoRFP AgentCore

  1. Integrate a Conversational Chatbot powered by Claude 3.5 Sonnet — enabling presales teams to ask natural-language questions about parsed RFPs, clarifications, or cost breakdowns.
  2. Add Amazon Q Business integration for knowledge retrieval from enterprise repositories.

Built With

Share this project:

Updates