Case StudyAI-native workflows

TwinMail - AI-native email workflows

ML-native email productivity workflows with background categorization, summarization, and offline archive ingestion.

Offline archive ingestion up to about 55 GB

Role

Founder and product architect

Team context

Founder-led build with architecture, product direction, and engineering execution coordinated through Twindevs.

Responsibility scope

Owned product framing, system architecture, and end-to-end implementation strategy.
Designed the ingestion, retrieval, and workflow-agent model for large mailbox archives.
Set the quality bar for grounded search, smart triage, and provider integration boundaries.

Stakeholders

Decision points

Keep ingestion tied to source archives rather than opaque derived-only stores.
Gate AI workflow features behind retrieval quality and source-traceable behavior.
Balance Google and Microsoft ecosystem integrations without coupling the product to one provider.

Problem and constraints

Build an AI-assisted email platform that can process large offline mailbox archives while preserving retrieval quality and user trust.

Architecture approach

Background agents for categorization, prioritization, summarization, and follow-up preparation.
Offline .mbox ingestion pipeline supporting archives up to about 55GB for semantic retrieval.
Feature-gated capabilities including AI search and smart views with provider integrations across Google and Microsoft ecosystems.

Outcomes

Established a practical ingestion and retrieval pattern for large mailbox datasets.
Reduced manual message triage overhead through ML-assisted prioritization.
Created a foundation for search and workflow automation tied to source data.

Next iteration

Expand retrieval evaluation coverage and release gates for accuracy and safe behavior before feature rollout.