case study · Logistic-WebApp
An end-to-end logistics operations platform - orders, warehouse, fleet, routes, pricing and billing. One unified platform instead of six separate systems. In production.
What it solves
Four pain points that follow every logistics company - and the architectural decisions that address them.
Before
Warehouse, shipping, billing - each in a different tool. Brittle integrations, unsynchronized data.
After
All modules on the same database, the same permission model, the same interface. From intake to delivery - a single flow.
Before
New field in a form, new report, new module - weeks of development.
After
163 JSON schemas define 45 entities. Adding a new entity is a config file, not server code.
Before
Customers call for updates, no self-service access to their inventory or documents.
After
Same platform, automatic data isolation. Customers see only their own data, down to the field level.
Before
Pricing, surcharges, driver scoring, monthly billing - hours of manual work, accumulated errors.
After
Flexible pricing per customer/zone/distance, automatic surcharges, monthly billing across multiple service categories.
Scope
Same database, same permission model, same interface. Every module lives in the context of the others - from warehouse intake to monthly billing, with no integration layers.
Full lifecycle: creation, warehouse picking, QR pallet scanning, controlled status transitions through to signed delivery.
Automatic intake, QR pallets, visual storage map, inventory tracking across 8 dimensions (stored/picking/ordered/picked/in-transit/available).
Vehicle registration, document expiry alerts (insurance/license/inspection), maintenance log, and performance-based driver scoring.
Multi-stop route optimization, weight-aware routing, digital signatures, live tracking. Interactive distribution zone management.
Multiple pricing models, per-customer/truck tariffs, automatic surcharges, monthly billing across multiple service categories.
Attendance, performance targets, automatic bonuses, employee documents. Automated monthly attendance reports.
Role-specific dashboards, revenue analytics, CSV export. Automated PDF reports with digital signatures.
What sets it apart
Maintenance documents, vehicle licenses and insurance are automatically parsed - dates, prices, types - with confidence scores and source citations. Less typing, fewer errors.
Same system, same DB, automatic data isolation. The customer sees only their own data - at the field level. No need to build a separate portal or maintain two codebases.
New entity, new field, new report - a JSON file. Changes that take weeks in other systems happen in hours. Less development, less technical debt.
Proof of capability
A system in production. Frontend in React/TypeScript. Backend in Node.js. Five internal services in Rust. An AI document-processing agent. External integrations. Cloud deployment. Every item - from decision to deploy - at the same workstation. No queues, no coordination, no waiting for someone else's answer.
A new system to build, an existing one to improve, or just an idea that needs to be translated into architecture - let's talk.
Let's talkArchitecture
Each service deployed in its own container. Failures are isolated, horizontal scaling is independent. Only the core server runs on Node - everything else on Rust, so PDF generation or route calculation never blocks the API event loop.
Schema engine, dynamic SQL builder, permission validation, RBAC model, and a Hooks system for business logic.
Gemini 2.5 Flash via Vertex AI. Returns JSON with a confidence score per field and a source citation from the original document.
Business document generation, digital signatures, Hebrew fonts and built-in RTL layout.
Geocoding, weight-aware truck routing, route optimization, synchronized street database.
Task scheduling, external data sync, maintenance alerts, and periodic automations.
Standalone audit log. Who, when, and what - before/after values for every action, with automatic archiving.
The stack
Core innovation
163 JSON schemas define 45 business entities. The schemas drive both the SQL layer and the UI simultaneously - the same single definition builds the dynamic query, the form, the table, and filters by permissions.
Fields, joins, rules, permissions, actions
Parameterized queries with table joins and aggregations
Tables, forms, and detail cards generated from the schema
Field-level access - sensitive data removed server-side
Hooks for custom logic. The standard CRUD pipeline is handled by the generic engine. For special cases - business automations, complex validations, or side effects - the Hooks system injects custom logic at defined lifecycle points, without touching the engine.
AI Docs · in production
A Rust service running on Gemini 2.5 Flash via Vertex AI. Input: PDF/JPEG of any business document - invoices, contracts, certificates, forms, or vehicle documents (production case study). Output: structured JSON - required fields, a confidence score per field, and a source citation from the original text for every extracted value.
Maintenance invoices / receipts
Licenses / insurance / inspection
Anti-hallucination. The prompt requires the model to cite the original text
for every returned value. When a document is unreadable, the field returns
{"status": "unreadable"}
instead of a guess. A manual correction interface is available for any output where the confidence score is low.
The service is also available as a standalone product for other offices - AI Docs page →
Automation
A scheduled task monitors an inbox, extracts data from messages in known formats, and creates or updates records in the database. A deduplication mechanism ensures each message is processed exactly once. Manual data entry that previously required a clerk becomes fully automatic.
defense in depth
Not a bolt-on module - security is built into each of 5 layers, and all thresholds are configurable from the management interface.
Adaptive rate-limiting, automatic IP blocking, brute-force protection, CORS, security headers.
JWT + MFA (TOTP/email), refresh token rotation, configurable lockout policy.
Categorical RBAC. Three access modes per permission: none / read / write. Field-level access per role.
Built-in multi-tenancy. Every query, view, and list is automatically filtered by customer.
Every action logged in an independent service - user, timestamp, before/after values. Automatic archiving.
heads-up
What happens when schema-driven architecture meets one developer and one year.
A new system architecture to design, an existing infrastructure to review, or a stack decision that needs a third eye.
Let's talk