Arryt — Courier Logistics v2

• What it is: a delivery/courier management system (used, among others, within the Restaurant chain (UZ) ecosystem — cf. check_arryt in crm_chopar). Admin panel for managers + REST/type-safe API + analytics + Flutter driver app + queues/cron. Order assignment, driver tracking, pricing engine, balances by terminal.
• Type / status / role: web-app (multi-app + mobile) / active / lead (per the case study "Lead Full-Stack Developer," team of 3; in git the user is the top author at 173/329 ≈ 53%, followed by Ilhombek 90, Shahzod 34).
• Activity period: 2023-11-04 → 2026-03-22 (~2.5 years); new_arryt is a modern rewrite (Elysia/Bun/Next15) of arryt (#8).

• API (`api/`): ElysiaJS (Bun) + Drizzle ORM (+typebox/zod) + PostgreSQL + Redis; BullMQ/Bull (jobs); DuckDB (@evan/duckdb — analytics); OpenTelemetry (@elysiajs/opentelemetry, auto-instrumentations, OTLP exporter — distributed tracing); server-timing, bearer-auth, Eden.
• Admin (`admin/`, `new_admin/`): Next.js 15 + React 19, Tanstack Query, shadcn/ui, real-time (WebSocket).
• Mobile (`mob/`): Flutter (Dart) — driver app: BLoC, GraphQL, ObjectBox (offline), Firebase (push), background GPS tracking.
• Other: mcp_server/ (custom MCP server!), protos/ + proto.ts (gRPC/protobuf), duck_api/ (DuckDB analytics), queue/, crons/, transfer_db/.
• DevOps: Docker, PM2, GitHub Actions, Nginx.

• Full rewrite on Elysia/Bun/Next15 (v2) with a type-safe API (Eden).
• DuckDB analytics on top of PostgreSQL (reports over 1M+ records).
• OpenTelemetry tracing for the API.
• Flutter driver app (offline mode, GPS, push).
• Custom MCP server (mcp_server/) — likely for AI access to logistics data.
• gRPC/protobuf layer (protos/, proto.ts).
• Prepared Arryt-Case-Study.md/html (portfolio case study) and arryt_monetization.md (monetization).
• Volume: 329 commits over ~2.5 years, 173 by the user (top author).

(From code/manifests + case study; the user is lead / top author.)

• DuckDB analytics on top of PostgreSQL (@evan/duckdb, duck_api/, order_data.duckdb): heavy analytical queries over 1M+ orders are offloaded to embedded-OLAP DuckDB, sparing the transactional Postgres. → A clean OLTP/OLAP split — an uncommon and strong choice.
• Distributed tracing (OpenTelemetry) in the Elysia API (auto-instrumentations + OTLP): production observability and latency optimization (<50ms per the case study). → A mature observability approach.
• Dynamic pricing engine: distance/time-based rules, driver bonuses, client-specific tariffs, daily-guarantee; rules cached in Redis + change audit.
• Real-time tracking of 500+ drivers: battery-efficient background GPS, geofencing for auto status changes, marker clustering on the map.
• gRPC/protobuf + custom MCP server: type-safe inter-service layer + AI access to logistics data.
• High-load processing: cursor pagination, bulk operations over 1000+ orders, index optimization.

• Sessions found: 39 .jsonl files in the local Claude Code sessions directory for this project. Plus .cursor / .cursorrules (26 KB) — Cursor + Claude Code.
• What was done with AI: v2 rewrite/features, likely case-study and MCP server generation. Not deeply sampled (large volume).
• Pattern: multi-tool (Claude Code + Cursor), custom MCP server as part of the product (AI access to data).