Goal
Reduce the mental overhead of email by turning a noisy inbox into a small number of clear, actionable decisions.
Traditional inbox management still requires repeatedly scanning messages, deciding what matters, remembering what needs attention and mentally tracking conversations that can wait. The objective was to remove as much of that work as possible while preserving confidence that important messages would not disappear into automation.
I designed a system that continuously reviews incoming email, understands conversations in context, separates genuine actions from noise and prepares useful replies where appropriate.
The aim was not to process more email. It was to make email require less attention.
Role
I designed, built and productionised the system end-to-end, combining deterministic workflow engineering with selective AI decision-making.
- Solution architecture. Designed the end-to-end email triage and response workflow around a simple, low-friction user experience.
- Pragmatic AI integration. Combined conventional automation with AI where interpretation or response generation added value, keeping the system predictable and controlled.
- Context-aware drafting. Built automated reply preparation that uses the available conversation context while keeping final sending under human control.
- Production reliability. Engineered the system for dependable recurring operation without duplicate work or continual manual cleanup.
- Extensible design. Structured the solution so additional email use cases can be introduced without compromising the core experience.
Capability
The system continuously reduces inbox clutter into a smaller, organised set of actions, prepared responses and messages worth attention — helping email demand less time and mental energy.
Business Impact
- Reduced inbox checking. Important email is surfaced automatically, reducing the need to repeatedly scan the inbox just to see whether something needs attention.
- Clear priorities. Incoming email is organised into simple action states, making it immediately clearer what needs a response, review or no immediate attention.
- Replies already waiting. Where enough context exists, useful reply drafts are prepared directly in the existing conversation, shortening the path from reading to responding.
- Smaller decisions. When human input is needed, the system surfaces exactly what requires attention while preparing the rest of the response.
- Less cognitive load. Low-priority messages are kept away from the active workload, reducing inbox noise and the feeling that everything needs attention.
- Better inbox UX. Consistent organisation turns the inbox into a clearer action interface rather than an undifferentiated stream of messages.
- Better searchability. Structured organisation makes past and current conversations easier to find without relying on remembering exact senders, subjects or keywords.
- Nothing important silently disappears. Automation is deliberately conservative, keeping important or uncertain messages visible rather than aggressively filtering them away.
- Human-controlled sending. Responses can be prepared automatically, while outbound email remains explicitly user-approved.
- Reliable ongoing operation. The system is designed to run repeatedly without creating duplicate work or requiring continual manual cleanup.
- Extensible automation. New email use cases can be added without redesigning the overall experience, allowing the system to evolve as requirements change.
Tech Stack
- Email & Messaging: Gmail ecosystem
- AI Layer: Large language model services
- Automation Layer: Workflow orchestration and event-driven automation
- Integration Layer: Email APIs and structured data interfaces
- Decisioning: Deterministic rules combined with AI classification
- State Management: Explicit workflow and conversation state
- Development: JavaScript / Python
- AI Engineering: Structured outputs, contextual classification, specialised prompting
- Operations: Scheduled processing, error handling, idempotency and workflow observability
Year
2026