Technical architecture, data flow, and maintenance reference for the Proposal Writer AI Agent.

1. System Overview

The Proposal Writer is an AI agent built in Make.com that automates project proposal drafting for an art fabrication workshop. It integrates three core platforms:

Platform Role
Make.com Orchestration layer — hosts the agent, tools, and automation scenarios
Airtable Database layer — stores Clients, Contacts, and Projects
Notion Output layer — proposal documents are generated here for human review

2. Architecture Diagram

flowchart TD
    A["📧 Dedicated Inbox"] -->|"New email"| B["Make.com Mailhook"]
    B --> C["AI Agent<br>(LLM + System Prompt)"]
    C --> D["🔍 Client Research<br>(Perplexity AI)"]
    C --> E["📋 Find or Create Client<br>(Airtable)"]
    C --> F["👤 Find or Create Contact<br>(Airtable)"]
    C --> G["📁 Create Project Record<br>(Airtable)"]
    C --> H["✏️ Update Project<br>(Airtable)"]
    C --> I["📄 Create Proposal<br>(Notion)"]

    D -->|"Research results"| C
    E -->|"client_id"| C
    F -->|"contact_id"| C
    G -->|"project_id"| C

    style C fill:#f9f,stroke:#333,stroke-width:2px

3. Data Flow

Workflow A — New Lead Intake

  1. Email received → Mailhook fires in Make.com
  2. Email parsed → Agent extracts client name, contact, brief, dates, budget signals
  3. Client research → Perplexity AI returns company facts
  4. Find or create client → Airtable searched by company_name; created if missing
  5. Find or create contact → Airtable searched by contact_name; created if missing
  6. Create project record → Minimal record created, linked to client and contact
  7. Draft proposal & update project → Agent populates scope, services, timeline, pricing, and written_proposal

Workflow B — Proposal Document Creation

  1. Trigger → Airtable project status set to "Create Proposal"