🤖 AI-Powered RFP Analyzer: An Azure-Based Multi-Agent Accelerator for RFP & Proposal Evaluation Using Semantic Kernel

A turnkey, multi-agent solution accelerator for automating the evaluation of RFPs/RFTs and Vendor/Supplier Proposals. This accelerator helps procurement teams reduce manual effort, ensure compliance, analyze vendor capabilities, and generate strategic negotiation recommendations—delivered as a structured, final evaluation report ; all deployed in your Azure tenant in under an hour.

---

🔍 Table of Contents

• ✨ Features
• 📐 Architecture
• ⚙️ Tech Stack
• ☁️ Provisioned Azure Services
• 🔧 Environment Variables
• 🚀 Step-by-Step Deployment
• 🤝 Contributing
• 📄 License

---

✨ Features

The AI-Powered RFP Analyzer combines modular AI agents, LLM orchestration with Semantic Kernel Agent Framework, and vector-based reasoning to deliver a complete procurement intelligence system. Each feature corresponds to a specialized agent that independently analyzes vendor/supplier proposals and contributes to a unified evaluation.

---

🧾 RFP/RFT Compliance Scoring Agent

Evaluate how well a Vendor or Supplier Proposal aligns with the original RFP/RFT document:

• Extracts and compares proposal content against mandatory RFP requirements.
• Identifies key gaps or misalignments in scope, deliverables, or terms.
• Assigns a compliance score (1–10) based on completeness and relevance.
• Highlights critical missing elements that may disqualify the vendor.

---

⚖️ Legal & Regulatory Risk Assessment Agent

Ensure the proposal meets internal policies and external regulatory standards:

• Cross-references legal clauses using an Azure-hosted vector index of procurement laws and policies.
• Flags missing or non-compliant sections such as liability, dispute resolution, or contract termination terms.
• Outputs a risk level: Low, Medium, or High, with a structured explanation.
• Helps legal and procurement teams pre-empt contractual risks.

---

🏢 Vendor Reputation & Stability Evaluation Agent

Assess the historical performance, financial stability, and credibility of vendors:

• Reviews past projects, industries served, customer satisfaction, and compliance history.
• Integrates with sample vendor vector indexes (or real data) for factual reputation signals.
• Produces a reputation score (1–10) reflecting reliability, experience, and maturity.
• Avoids speculative or price-driven assessments—focuses purely on track record and signals.

---

🌍 Market Intelligence Integration Agent

Incorporate contextual industry insights to inform strategic decision-making:

• Simulates or integrates with APIs to gather market conditions.
• Benchmarks the vendor against competitors, highlighting differentiators or red flags.
• Evaluates macro risks such as economic conditions, supply chain volatility, or regulatory trends.
• Outputs a market risk rating (Low, Medium, High) with supporting narrative.

---

🧠 Negotiation Strategy Generator Agent

Craft actionable negotiation strategies tailored to each vendor:

• Synthesizes insights from compliance, legal, vendor, and market agents.
• Identifies leverage points (e.g. gaps, dependencies, differentiators).
• Recommends a strategy type: Defensive, Balanced, or Aggressive.
• Provides structured suggestions for mitigation, incentives, and fallback positions.

---

📊 Final Evaluation Report Compilation Agent

Consolidate all agent outputs into a clear, concise, and decision-ready report:

• Aggregates findings into a unified assessment with supporting justifications.
• Assigns a final vendor score (1–10) indicating overall suitability and risk level.
• Includes highlights, concerns, and strategic recommendations.
• Outputs a Markdown or PDF document for stakeholder circulation.

---

Together, these features allow procurement teams to move faster, reduce subjectivity, ensure compliance, and drive stronger vendor outcomes with confidence.

📐 Architecture

🔹 High-Level Infrastructure

• User Interaction: End users interact via a web browser to upload documents and trigger evaluations.
• Azure Container Apps Environment: Hosts the core containerized Python application.
• Worker Container App: Runs the multi-agent logic for evaluation and reporting.
• Connected Azure Services:
  • Azure OpenAI – Powers the intelligent agents using GPT-based models.
  • Azure AI Search – Enables semantic and vector-based retrieval of indexed data.
  • Azure Document Intelligence – Extracts structured data from RFPs and proposals.
  • Azure Storage Account – Stores document inputs and intermediate outputs.
  • Azure Container Registry (ACR) – Hosts the application’s Docker image.
  • Azure Storage Account – Stores uploaded documents and temporary data.

---

🔹 Solution Workflow

1. 📤 Upload
   Users upload RFPs and Vendor/Supplier Proposals via the web interface or API.

2. 📄 Document Parsing & Summarization
   The system extracts and summarizes content from both documents using Azure Document Intelligence.

3. 🤖 Multi-Agent Evaluation
   The Semantic Kernel Orchestrator initiates an agent group chat with specialized agents:

   • RFP Compliance Agent: Compares the proposal against RFP requirements.
   • Legal Compliance Agent: Identifies legal and regulatory risks.
   • Vendor Evaluation Agent: Assesses vendor reputation, stability, and performance.
   • Market Intelligence Agent: Evaluates market trends and competitive risks.
   • Negotiation Strategy Agent: Recommends negotiation approaches based on findings.

4. 📊 Report Generation
   The Evaluation Report Generator Agent consolidates all agent assessments into a final, structured Vendor Proposal Evaluation Report, including scores, risks, and recommendations.

---

⚙️ Tech Stack

• Language: Python 3.12.10
• Containerization: Docker
• Infrastructure-as-Code: Bicep (located in /infra folder)
• Deployment Target: Azure Container Apps
• Source Code: Located in /src folder
• Image Registry: Azure Container Registry
• Multi-Agent Orchestration: Semantic Kernel
• LLM Integration: Azure OpenAI
• Document Parsing: Azure Document Intelligence
• Search/Indexing: Azure AI Search (Vector + Semantic)

---

☁️ Provisioned Azure Services

The Bicep templates provision the following Azure resources:

• Azure Container Apps – Hosts the containerized multi-agent Python app.
• Azure Container Apps Environment – Provides isolated execution for the app.
• Azure Container Registry (ACR) – Stores and serves the app container image.
• Azure Document Intelligence – Extracts structured data from RFP documents.
• Azure OpenAI – Powers intelligent agents using GPT-based models.
• Azure AI Search – Enables semantic and vector-based search across processed content.
• Azure Storage Account – Stores uploaded documents and temporary data.

---

🔧 Environment Variables

Environment variables are used to configure service connections and runtime behavior. These are typically managed by AZD and include:

• AZURE_OPENAI_ENDPOINT
• AZURE_OPENAI_KEY
• AZURE_SEARCH_SERVICE_NAME
• AZURE_STORAGE_ACCOUNT
• AZURE_FORM_RECOGNIZER_ENDPOINT

You can set or review these in .env or via the azd env commands.

---

🚀 Step-by-Step Deployment

Follow these steps to deploy the solution using AZD:

1. Prerequisites

• Azure Developer CLI (AZD)
• Docker Desktop
• Azure CLI
• The accelerator expects two vector-indexes created on Azure AI Search to facilitate the run of Legal Compliance Agent and Vendor Evaluation Agent, named legal-policy-index and supplier-insights-index respectively. Please customize and create them per your use-case data or use the sample documents available under src/documents/sample-docs/index-creation.

🚀 Step-by-Step Deployment

Follow the instructions below to deploy the AI-Powered RFP Analyzer into your Azure environment using Azure Developer CLI (AZD). This guide ensures a clean and repeatable deployment flow.

---

🔹 1. Clone the Repository

Start by cloning the repository to your local machine:

git clone https://github.com/aadrikasingh/AI-Powered-RFP-Analyzer.git
cd AI-Powered-RFP-Analyzer

---

🔹 2. Log in to Azure

Ensure you're authenticated with your Azure account via the Azure CLI:

az login

This will open a browser window prompting you to sign in with your Azure credentials.

✨ Note: If the browser does not open automatically, copy the provided URL and paste it into your browser manually. Then enter the one-time code shown in your terminal.

Once authenticated, your terminal will confirm your login and list the available subscriptions.

---

🔹 3. Set Up Environment Variables

Before deploying, configure the required environment variables.

📄 Sample file location:src/.env.sample

📅 Create your working .env file:

1. you will find .env.sample in src/ directory
2. copy it inside src/src/ and rename it to .env
3. you can edit .env you created with your specific needed values as shown below.

✅ Mandatory Environment Variables

Variable Name	Description
AZURE_AI_SEARCH_ENDPOINT	The endpoint URL of your Azure AI Search instance
AZURE_AI_SEARCH_API_KEY	Admin or query key for your Azure AI Search instance
LEGAL_POLICY_INDEX	Name of the index containing legal/policy documents (e.g., legal-policy-index)
SUPPLIER_INDEX	Name of the supplier insights index (e.g., supplier-insights-index)

💡 You can use the provided default values in .env.sample if you have matching indexes already created in your Azure AI Search resource.

---

🔹 4. Initialize and Deploy the Environment

Use the Azure Developer CLI to initialize and deploy your solution.

🛠️ Initialize the AZD Environment

azd init

You will be prompted to enter an environment name. This name will be used as a prefix for all provisioned Azure resources.

📌 Example:> If you enter dev, your container app name might look like ca-dev-hlukwx2gjy4lo, where:

• ca = Container App
• dev = Environment name
• hlukwx2gjy4lo = Random string for uniqueness

---

🚀 Deploy the Solution

azd up

You will be prompted to:

1. Select an Azure Subscription to use for deployment.
2. Select a deployment region.

🌍 Recommended Region: Use a region that supports all required services, such as Sweden Central.

Once confirmed, AZD will begin provisioning the infrastructure using Bicep templates. This includes:

• Azure Container Apps
• Azure OpenAI (OpenAI will provision two models: text-embedding-ada-002 and gpt-4o. Both models will have 10,000 TPM. If you want to change this value, go to the OpenAI Bicep module and decrease the capacity number in the model's SKU configuration)
• Azure AI Search (if you are using an exisiting Ai search service then you can remove the Ai search provioned thourgh bicep template)
• Azure Document Intelligence
• Azure Container Registry (ACR)
• Monitoring (Log Analytics & Application Insights)

⏱️ Provisioning typically takes 5–10 minutes, depending on your region and resource availability.

✅ Once the AZD template deployment completes, a URL will be displayed in your terminal output. You can use this URL to launch and explore the AI-Powered RFP Analyzer web application in your browser.

🔹 4. Removing the solution

Run azd down to remove the Infra and app provioned throgh this template.

azd down

🤝 Contributing

We welcome contributions to enhance and evolve the AI-Powered RFP Analyzer accelerator. To get started:

• Fork the repo
• Create a feature branch
• Make your changes
• Push and open a PR

📄 License

This project is licensed under the MIT License.

⚠️ Note: This solution is intended for internal Microsoft use as an accelerator for proofs of concept (POCs) and demo environments. It is not a production-ready offering. Use at your own discretion.