← Back to Case Studies Multi-Agent AI

Propose-AI: Autonomous Proposal Generation

Multi-agent AI system that generates compliant first-draft proposals from RFP documents using AWS serverless architecture and RAG-based knowledge retrieval.

The Challenge

Government contractors spend enormous time and resources responding to RFPs:

Proposal teams spend 40-80 hours per RFP extracting requirements and drafting responses
Compliance gaps are often discovered late, requiring expensive rework
Institutional knowledge from past winning proposals is underutilized
Quality varies significantly based on team experience and availability

The Solution

Propose-AI uses a multi-agent architecture where specialized AI agents collaborate to produce compliant proposal drafts:

┌─────────────────┐
│  Orchestrator   │  (State Machine Controller)
└────────┬────────┘
         │
    ┌────▼────┐
    │ Shredder│  → Structured JSON (requirements extraction)
    └────┬────┘
         │
    ┌────▼─────────┐
    │  Strategist  │  → Draft Proposal (uses RAG)
    └────┬─────────┘
         │
    ┌────▼────┐
    │   QC    │  → Compliance Report
    └────┬────┘
         │
    [Score < 95%?] ──Yes──→ Retry (max 3x)
         │
         No → END

RFP Shredder Agent — Extracts requirements, deadlines, and scoring criteria from PDF/DOCX
Content Strategist Agent — Generates proposal content using RAG to retrieve winning content
Compliance & QC Agent — Validates draft against requirements, flags gaps, scores compliance
Orchestrator — Coordinates workflow with automatic retry logic

Technology Stack

AWS Lambda AWS Bedrock Claude 3 Opus Pinecone API Gateway Aurora Serverless S3 Electron React

Key Features

Autonomous Operation — No human intervention required after RFP upload
RAG-Powered Memory — Retrieves relevant content from past winning proposals
Iterative Refinement — Auto-retries until compliance target achieved
Desktop Application — Kiro-style interface with file explorer, markdown preview, and AI chat
Type-Safe Schemas — Pydantic models ensure structured, validated outputs

Desktop Application

The Electron-based desktop app provides a familiar IDE-style interface:

File Explorer (Left Panel) — Browse local files and SharePoint documents
Markdown Preview (Top Right) — Live preview of generated proposal sections
AI Chat (Bottom Right) — Conversational interface for proposal generation

Security

All data encrypted at rest (S3, Aurora) and in transit (TLS)
API keys stored in AWS Secrets Manager
VPC isolation for database layer
IAM least-privilege policies
CloudTrail logging for audit trail