Case Study

Stethy: AI Email Triage

Backend, infrastructure and tooling for an invite-only B2B email triage SaaS

Yash Soni · 2024 to present

The Problem

Stethy needed an email triage workflow platform that enterprise users can connect to their inbox.

The Solution

Stethy is an invite-only B2B SaaS that connects to a mailbox, classifies incoming email and routes it through defined workflows. A deterministic, versioned rules engine does the classification first, with AI triage and confidence scores alongside.

Architecture & Tech Stack

• Backend: Python, FastAPI, SQLAlchemy, Alembic
• Frontend: TypeScript, React, Vite
• Database: PostgreSQL
• Mail: Microsoft Outlook and Azure AD, Gmail extension
• LLM gateway: LiteLLM
• Infrastructure: AWS CDK (Python), Lambda, CodePipeline, multi-account dev, stage and prod
• Observability: Langfuse on Azure Container Apps

Key Features

• Rules Engine: Versioned classification and routing rules with a dry-run to test an email before it goes live
• Connectors: Outlook mailboxes, including bring-your-own app credentials, and a Gmail extension
• Runs and Audit Trail: A live execution log and a classification history with confidence, matched rule and manual correction
• Multi-Tenant Org Management: Workspaces, users, invites, usage and access logs

My Role

I work with Stethy, a client based in Australia, as a contractor across backend, frontend and infrastructure.

Last updated: 2026-10-11