Automart

A dealership operations platform with a streaming AI assistant, a DB-backed 65-permission authorization engine, and a resilient Redis caching layer.

🚧Internal Services Demo Coming Soon!

LIVE

PHOTO

Overview

Automart handles the operational data model behind an automotive dealership: staff, sessions, customers, vehicles, messages, announcements, inventory, notes, permissions, and reports. Built with Next.js 16 and React 19, it uses a hybrid authorization model instead of rigid role-only gates, keeps Redis optional through racing timeouts and PostgreSQL fallbacks, and protects multi-step writes with isolated database transactions.

Technical Stack

Next.js 16 (App Router)

React 19

TypeScript

PostgreSQL (Neon) & Drizzle ORM

Key Features

Server-defined tools for staff, inventory, customer, message, notification, web-search, code-execution, and reporting queries via AI SDK 7 with Gemini, Vertex AI, and DeepSeek provider routing.
Dynamic SQL report builder with validated joins, aggregations, permission checks, and field-level redaction for sensitive data.
Queries execute through tightly-scoped server tools with sandboxed database access, keeping model output away from direct database execution.
Streams markdown tokens, live tool outputs, and structured source references with full stop-generation capability.

Architectural Highlights

Designed a resilient cache-first architecture utilizing Upstash Redis with 250ms racing timeouts, a 60s circuit breaker, and PostgreSQL fallbacks.
Engineered a tool-based streaming AI architecture leveraging sandboxed server tools and field-level permission redaction for safe data access.
Developed a custom hybrid authorization system handling a 65-permission catalog, role defaults, and per-user allow/deny overrides.
Expanded the Vitest coverage to 662 cases across 168 files covering permissions, routes, caching, table flows, AI tools, and auth behavior.
Built a unified, high-performance data table engine that drives 9 application views with normalized URL paging and debounced filtering.

Project Status

Robust hybrid authorization engine and matrix-view security control panel.
Tool-based AI conversational companion with sandboxed server-tool database access.
Resilient circuit-breaking cache architecture protecting stateful session layers.
Bilingual announcement workflows, CRM customer registries, and VIN-decoding vehicle engines.
Current Vitest suite tracks 662 tests across 168 test files alongside strict no-emit TypeScript checks.