Skip to content

Architecture Blueprint: Ticket Masala

This document provides a comprehensive deep-dive into the technical architecture of Ticket Masala, from its core design philosophy to its specific data models and AI integration patterns.

Design Philosophy: The Modular Monolith

Ticket Masala is designed as a Modular Monolith following a "Monolith First" doctrine. This approach prioritizes operational simplicity and high performance without the overhead of microservices, while maintaining strict logical boundaries between subsystems.

Logical vs. Physical Separation

We utilize standard .NET abstractions to achieve logical separation while running in a single process: - In-Memory Messaging: System.Threading.Channels instead of RabbitMQ. - Local Persistence: SQLite (WAL Mode) with FTS5 for high-speed indexing. - Background Processing: IHostedService instead of separate worker containers.

The Universal Entity Model (UEM)

To remain domain-agnostic, Ticket Masala operates on three core abstractions. terminologies are mapped at the presentation layer via configuration.

UEM Concept Internal Implementation Description
WorkItem Ticket The atomic unit of work.
WorkContainer Project A high-level grouping of work items.
WorkHandler ApplicationUser The entity (human or system) resolving the work.

Intelligence: The GERDA AI Engine

The GERDA engine (Grouping, Evaluation, Ranking, Dispatching, Anticipation) orchestrates the AI-driven workflow optimization.

AI Processing Pipeline

  1. Ingestion: Data enters via Gatekeeper API or SAP Sync.
  2. Classification (Local): ML.NET classifies the work item and estimates effort points.
  3. Enrichment (Ephemeral): OCR and PII scrubbing are performed in-memory.
  4. Ranking (WSJF): Weighted Shortest Job First algorithm scores priority.
  5. Dispatching: GERDA recommends the best handler based on affinity, skill, and load.

Data Architecture

Hybrid Storage Model

We use a hybrid relational and document-based storage approach in SQLite: - Universal Fields: Indexed relational columns for core properties (Status, CreatedAt, DomainId). - JSON Fields: A CustomFieldsJson column stores domain-specific data without requiring schema migrations.

Performance Optimizations

  • WAL Mode: Enables concurrent reads and writes.
  • FTS5: Full-text search engine integrated for blazingly fast lookups across tickets and the knowledge base.
  • Computed Columns: Key JSON properties are extracted into virtual columns for indexing.

Security & Governance

The Identity Layer

  • RBAC: Fine-grained Role-Based Access Control (Admins, Team Leads, Agents, Customers).
  • Isolation: Strict data partitioning via TenantId and DomainId.

AI Sovereignty

  • PII Proxy: Local scrubbing before any third-party AI service is called.
  • Budget Caps: Hard and soft tokens caps per tenant to prevent operational shock.

Deployment Strategy

Ticket Masala is optimized for Cloud-Local operation: - Docker-Ready: Chiseled container images for minimal attack surface. - Single-Container: One container contains the Web UI, API, and background AI workers. - Fly.io Optimized: Ready for global distribution with local persistent volumes.