InboxIQ

InboxIQ is an end-to-end email intelligence product that takes users from authentication and Gmail OAuth synchronization through LLM triage and a dashboard.

I shipped the system to production on Next.js and Vercel with a FastAPI backend on Railway, delivered all 11/11 MVP acceptance criteria, and closed six P0 blockers after midpoint review.

Email automation combines sensitive OAuth credentials, tenant-isolated data, asynchronous ingestion, and probabilistic model output. The architecture needed strong security boundaries while keeping synchronization and triage work off the interactive request path.

I designed and shipped the full product flow across the Next.js frontend, FastAPI services, PostgreSQL data model, Gmail integration, background workers, and deployment infrastructure.

Gmail OAuth 2.0 tokens are stored server-side only behind JWT-authenticated FastAPI endpoints. Supabase PostgreSQL row-level security provides the primary tenant-isolation boundary across 20 migrations.

Background synchronization and triage workers process new mail outside the request path. The LLM pipeline performs classification, summarization, and structured extraction with Pydantic JSON-schema validation plus provider failover and retry logic.

The hardest part was joining OAuth security, row-level data isolation, background processing, and fallible model output into one reliable product flow. An automated evaluation harness and strict structured-output validation kept triage quality measurable, holding ≥85% accuracy on test fixtures.

InboxIQ reinforced that an AI feature is only one part of a production system. Credential handling, tenant isolation, migrations, background execution, schema validation, retries, evaluation, and deployment all determine whether the model is actually useful.