TMAll workRésumé PDF ↓

2026 · Demo prototype / AI automation

Foreman

AI operations manager prototype running on synthetic construction data.

Foreman is an interactive demo prototype developed for Allsaints Builders to explore AI-assisted email management. Note: This is an evaluation prototype operating strictly on synthetic mock data, not a live production deployment connected to active contractor inboxes.

Foreman screenshot — 1 of 3
Foreman screenshot — 2 of 3
Foreman screenshot — 3 of 3
1 / 3

What it does

  • Evaluation prototype on synthetic data: This project is an interactive demo prototype, not an active production deployment connected to live contractor inboxes. The entire system, including dual monitored inboxes, client inquiries, supplier quotes, and audit logs, operates entirely on synthetic test scenarios designed to validate triage accuracy and response workflows without handling actual client communications.
  • Live prototype access & credentials: A public deployment of the Next.js review dashboard is available at the Demo link above. Evaluators can sign in using test credentials: Email: 123@gmail.com | Password: admin123. The dashboard allows exploring the live triage feed, urgency badges, draft inspection, and simulated staff approval flow.
  • Dual-inbox operational triage: The architecture models the workflow of Allsaints Builders, a UK construction firm managing two high-volume Titan email streams: info@ for commercial quotes and director administration, and team@ for subcontractor updates and site progress. The prototype demonstrates how high-priority messages can be surfaced immediately while routine updates are organized.
  • Nine-theme classification & drafting: An automated n8n pipeline extracts incoming RFC 822 email metadata and utilizes OpenAI GPT-4o to categorize messages across nine construction themes (including orders, accounts, finance, suppliers, NPH jobs, appointments, invoices, and certificates). For actionable messages, GPT-4o drafts responses with strict [CONFIRM ...] safety placeholders for unverified details.
  • Mobile voice review & WhatsApp alerts: For messages flagged as urgent, database triggers simulate Twilio WhatsApp notifications containing secure 32-byte HMAC-SHA256 tokens. A dedicated mobile voice portal enables hands-free review via Whisper STT and voice synthesis, enforcing an intentional physical touchscreen confirmation before any simulated dispatch.
  • Decoupled architecture & audit log: Supabase PostgreSQL 15 acts as the single source of truth between the n8n automation engine and the review UI. An immutable audit trail records every classification, AI draft, edit, and approval action with actor attribution. Once approved, records enter a locked state to prevent duplicate submissions or race conditions.