VWare

A .NET / C# marketplace finance & pricing-automation platform for Iranian marketplaces (SnappShop, TapsiShop) — supplier price crawling, an automated pricing engine, inventory reconciliation, and multi-channel seller operations.

---

What it is

VWare automates the finance and pricing side of running a marketplace reselling business. It continuously crawls supplier prices, computes floor prices and margins through a pricing engine, reconciles marketplace listings and inventory against live supplier availability, and pushes updates across multiple sales channels — with an operations dashboard on top.

It is not a demo: VWare powered a live phone-reselling operation on SnappShop (a Snapp Group marketplace), where pricing and reconciliation were fully automated by the platform.

Architecture

VWare is a layered .NET 9 solution of 37 projects built on a small in-house application framework — GiliX — rather than a third-party stack, so the CQRS/mediator pipeline, logging, and cross-cutting behavior are owned and testable end to end.

flowchart TB
    subgraph CH["Marketplace channels"]
        SS["SnappShop"]
        TS["TapsiShop"]
    end
    subgraph ING["Price ingestion"]
        CRAWL["Crawler Service<br/>supplier price crawling + retry/backoff"]
        LINK["AI Link Filler<br/>Claude-assisted product matching"]
    end
    subgraph CORE["Pricing &amp; reconciliation engine"]
        CALC["Calculator / Viravin.Calculator<br/>floor-price + margin computation"]
        RECON["Reconciliation<br/>smart inventory reset"]
    end
    subgraph OPS["Multi-channel operations"]
        AUTO["SnappShop Automation<br/>price / stock / listing sync"]
        TIC["TapsiShop ItemCreator"]
        BOT["Telegram Bot / SMS Reader"]
    end
    subgraph FW["GiliX framework + domain (foundation)"]
        MED["GiliX.Mediator — CQRS"]
        DOM["Domains + Persistence — EF Core"]
        DIAG["VWare.Diagnostics — secret-safe logging"]
    end
    DASH["VWare.Dashboard — operations UI"]

    SS --> CRAWL
    TS --> CRAWL
    CRAWL --> CALC
    LINK --> CALC
    CALC --> RECON
    RECON --> AUTO
    RECON --> TIC
    AUTO --> SS
    TIC --> TS
    CALC --> MED
    AUTO --> MED
    MED --> DOM
    CALC --> DIAG
    DASH --> DOM

Loading

Why this design

• Custom mediator (GiliX.Mediator). Commands, queries, and events flow through one owned CQRS pipeline with a ValidationBehavior seam. Owning the mediator (instead of pulling in a third-party one) keeps the request pipeline small, dependency-free, and fully unit-testable at the boundary.
• Domain layering. Domains holds the domain model; Persistence isolates EF Core. Business services and channel clients depend on the domain, not on infrastructure — so a marketplace or storage change stays contained.
• Reconciliation as a first-class flow. The pricing run does a smart inventory reset: the first pass resets all listings, and subsequent passes only reset products that lost a valid supplier price — so a product removed from the source, or one whose supplier price can't be found, is never left purchasable at a stale price. This is the core "don't lose money on stale listings" guarantee.
• Secret-safe logging. VWare.Diagnostics.RedactionEngine scrubs tokens, passwords, and PII out of logs and diagnostic bundles before they are ever written.

Solution structure

Layer	Projects	Responsibility
Framework (GiliX)	GiliX, GiliX.Common, GiliX.Logging, GiliX.Mediator	CQRS/mediator pipeline, logging, shared primitives
Domain	Domains, Persistence	Domain model + EF Core persistence
Services	Authentication, Calculator, Crawler, SnappShopAdmin	Core business services
Channel clients	VWare.SnappShop.Service / .Client / .Automation, Viravin.TapsiShop.Service, TapsiShop.ItemCreator, VWare.Crawler.Service, VWare.Calculator.Service, VWare.TelegramBot, VWare.SmsReader	Marketplace integrations & runnable services
Pricing	Viravin.Calculator	Floor-price & margin engine
Operations UI	VWare.Dashboard	Monitoring / operations surface
Shared	Shared/VWare.Diagnostics	Secret-safe logging, health checks, diagnostics
AI	VWare.AI	Claude (Anthropic) integration for product matching
Tests	*.Tests (Crawler, Calculator, SnappShop Automation, TapsiShop, Diagnostics)	xUnit unit & integration suites

Engineering rigor

• Compile-time "Time Policy". BannedSymbols.txt + Roslyn BannedApiAnalyzers turn DateTime.Now, DateTime.Today, Persian-calendar output, and ambiguous DateTime.Parse into build errors — forcing UTC / IAppClock everywhere. Server-timezone drift is a real correctness hazard for pricing and reconciliation, and this makes it unrepresentable.
• Directory.Build.props — nullable reference types, implicit usings, .NET analyzers at latest, and warnings-as-errors for banned APIs, applied solution-wide.
• .editorconfig — consistent formatting and style across every project.
• Secret-safe logging — RedactionEngine (with its own test suite) keeps credentials and PII out of logs.
• Dedicated test projects — xUnit suites covering the crawler, calculator, SnappShop automation, TapsiShop pricing, and diagnostics.
• GitHub Actions — automated code-review and release-publishing workflows.

Operating impact

VWare priced and reconciled a ~10,000-SKU catalog (5,000+ active at a time) across 15 suppliers for a live phone-reselling operation on SnappShop (a Snapp Group marketplace), cutting manual reconciliation/listing work by ~60% through transaction-level mapping and reusable pricing logic. In a documented ~2-week window, with pricing and reconciliation fully automated by VWare:

• ~673M Toman gross merchandise value (GMV)
• ~32.5M Toman net profit across 67 orders
• ~5% blended net margin after platform fees (which reach up to ~20% per SKU)
• sourcing from 15 suppliers

How this was built

VWare was developed spec-driven and AI-agent-orchestrated (see AGENTS.md): the human owns the architecture, the domain model, and the trade-offs; agents execute against tightly-scoped specifications and a strict developer contract — a single runtime crawl pipeline, a single per-shop config source, and structured, correlation-id crawl logs. The emphasis is ownership and verifiability, not automation for its own sake.

Getting started

Requires the .NET 9 SDK.

dotnet restore Viravin.sln
dotnet build   Viravin.sln -c Release
dotnet test    Viravin.sln

Configuration

Real appsettings*.json files and any credentials/ are gitignored and never committed. Copy the template and fill it from your own environment:

cp appsettings.example.json appsettings.json

Then set: ConnectionStrings:DefaultConnection, Snappshop:Token, Telegram:BotToken, GoogleSheets:CredentialsPath, and OpenRouter:ApiKey.

License

Released under the MIT License. © 2026 Mahdi Aghakhani.