# Document Management System

## 1. Introduction and Objectives

The goal is to develop a lean, server-based document management system focused on implementing an interface first rather than a frontend. Initial usage will be entirely programmatic via an API; a frontend is intentionally scheduled for a later expansion phase.

The system shall process PDF documents, capture their contents via OCR, provide full-text search capabilities, and enable LLM-supported content enrichment. Additionally, automated file import from a monitored folder shall be supported.

Key quality objectives:
- Simple and lean solution.
- Minimal frameworks and external systems; external components only where they provide clear added value.
- Testable architecture with comprehensive unit test support.
- Traceability and consistency of state changes through Event Sourcing.
- Clear, maintainable, and extensible structure.
- Simple installation, operation, and automation.
- Reliable import and consistent processing.
- Reliable searchability of documents.

## 2. Constraints and Boundary Conditions

- The implementation language is Go.
- The architecture shall be described in a technology-agnostic way to avoid unnecessary constraints on specific tools or products.
- The system shall initially be usable without a frontend; access is provided via the API.
- Automation for build, test, and deployment shall consider the existing Gitea and Actions environment.
- Documentation shall follow the arc42 structure.
- Unit tests shall be provided for all essential architectural components and behaviors.

## 3. Context and Scope

### Business Context

The system manages incoming documents, particularly PDF files. Documents can be actively submitted via the API or automatically ingested from a defined import folder. Later, it shall also be possible to send documents via email to the system for import.

After import, documents are processed, analyzed, made searchable, and transferred to a final storage structure. Processing also includes enrichment with additional information from LLM-supported analysis.

### Technical Context

External or adjacent systems and interfaces:
- File system for import folder, temporary processing, and final storage structure.
- API clients for using the system interface.
- LLM integration for semantic analysis, tagging, or similar content processing.
- OCR processing for text recognition from documents.
- Automation environment with Gitea and Actions for build, test, and deployment processes.

## 4. Solution Strategy

The solution shall be implemented incrementally and with minimal complexity:

1. First, the backend with a clear API shall be implemented.
2. Then import, parsing, OCR, search indexing, and event processing shall be integrated.
3. A frontend is intentionally deferred and considered only after stabilization of the interface.

Core principles of the solution:
- API-first approach.
- Lean, modularly designed components.
- Clear separation of domain logic, infrastructure, and application logic.
- Event Sourcing for traceability of changes and support of data consistency.
- Testability as an architectural guiding principle, not an afterthought.
- Automated file-based import as an integral part of the solution.

## 5. Building Block View

### 5.1 Overview of Main Components

The system consists of the following central building blocks:

- API Layer
- Application Logic
- Document Management
- Import and Watcher Component
- Processing Pipeline
- OCR Component
- LLM Integration Component
- Full-Text Search Component
- Event and Consistency Component
- Storage and Structuring Logic
- Test Layer

### 5.2 Building Block Descriptions

#### API Layer
Provides the interface through which documents can be imported, searched, queried, and managed. Since the frontend is not initially implemented, the API is the primary entry point of the system.

#### Application Logic
Coordinates use cases such as import, processing, search, relocation, and retrieval. This layer encapsulates workflow logic and shall be independently testable from infrastructure components.

#### Document Management
Manages metadata, status, identities, and the business lifecycle of a document. This includes states such as newly imported, in processing, processed, archived, or failed.

#### Import and Watcher Component
Monitors a defined folder for newly deposited documents. When a document is saved there, automatic processing shall be triggered.

Additional requirements:
- Documents from the import folder are automatically detected.
- Imported files can receive a timestamp in the filename documenting the import time.
- Automatic processing starts without manual intervention.

#### Processing Pipeline
Responsible for the overall flow of a document after ingestion. This includes reading, parsing, OCR, content enrichment, search indexing, status changes, and final storage.

#### OCR Component
Extracts text from imported PDF documents where text is not directly or fully machine-readable. OCR results serve as a foundation for search and further analysis.

#### LLM Integration Component
Enables content analysis of documents through an LLM. The goal is particularly semantic enrichment, such as through tagging, classification, structuring, or similar additional information.

#### Full-Text Search Component
Provides full-text search over processed document contents. This is based on extracted and potentially OCR-derived text content as well as relevant metadata.

#### Event and Consistency Component
Implements Event Sourcing at the business level. Relevant state changes in documents are recorded as events so that changes remain traceable and state can be consistently reconstructed.

Important requirements:
- No commitment to specific technologies in this documentation.
- Events shall be expressed in business-understandable terms.
- The sequence of events shall support auditability and recoverability.

#### Storage and Structuring Logic
Processed documents are transferred from the import context to the final storage structure. The following requirements shall be considered:

- Processed documents shall not remain in the import folder.
- Processed documents shall be transferred to a processed target area.
- The final folder structure shall reflect actual storage.
- Imported and processed files shall be moved to year-based folders.
- The filename can be supplemented with an import timestamp.

#### Test Layer
All relevant layers and components shall be secured through unit tests. The architecture shall be designed so that business logic remains independently testable and infrastructure dependencies are interchangeable or encapsulated.

## 6. Runtime View

### 6.1 Import via Monitored Folder

1. A document is saved in the defined import folder.
2. The watcher component detects the new document.
3. The system takes over the document for processing.
4. It is parsed and, if necessary, made textually accessible via OCR.
5. The LLM enriches the document semantically.
6. Contents are indexed for full-text search.
7. Business-relevant steps are captured as events.
8. The document is moved to the final storage structure.
9. The file is sorted into a year-based folder.
10. Optionally, the filename receives an import timestamp.

### 6.2 Import via API

1. A client submits a document via the API.
2. The system creates the document and starts processing.
3. Subsequently, the same processing flow as file-based import is followed.
4. The document is finally made searchable and archived.

### 6.3 Search

1. A client sends a search query to the API.
2. Full-text search queries indexed document contents and relevant metadata.
3. The system returns matching results.

### 6.4 Traceability of State Changes

1. Each significant action on a document generates a business event.
2. These events form the foundation for traceability and consistency.
3. The status of a document can be derived or reconstructed from recorded events.

## 7. Deployment View

The system is intended for operation on a personal or dedicated server. It shall require minimal external dependencies and be deployable in a simple operating environment.

Essential deployment view:
- Backend service for providing the API and orchestrating processing.
- Access to file system areas for import, processing, and final storage.
- Integration of required components for OCR, search, and LLM usage.
- Build, test, and deployment automation via Gitea and Actions.

## 8. Cross-Cutting Concepts

### 8.1 Testability

Testability is a central architectural principle. The architecture shall be designed so that core business logic can be tested independently of infrastructure.

Requirements:
- Unit tests for all essential modules.
- Clear interfaces between business logic and infrastructure.
- Good mocking or stubbing capability for external dependencies.
- Reproducible tests for import, processing, search indexing, and event processing.

### 8.2 Event-Orientation and Consistency

Business state consistency shall be supported through an Event Sourcing concept. Changes to documents are considered not only as current state but also as a traceable sequence of business events.

### 8.3 File Import and Document Lifecycle

Documents can be actively imported or passively ingested from a monitored folder. After successful processing, they shall be orderly transferred to the final folder structure.

Rules:
- Automatic processing when saved to the import folder.
- Marking or renaming with import timestamp is possible.
- Organization by year in the target storage.

### 8.4 Search Concept

Full-text search is an integral part of the system. It shall ensure discoverability through document contents and consider both directly readable and OCR-derived contents.

### 8.5 LLM-Supported Enrichment

LLM integration serves semantic processing of documents. This can include tagging, content structuring, classification, or similar forms of enrichment.

### 8.6 API-First

The roadmap provides for initially making the interface exclusively operationally usable. A frontend is intentionally deferred so that core functionality can be built independently and stably.

### 8.7 Documentation

Architecture and system documentation shall be maintained in an arc42-oriented form. The documentation shall clearly depict essential decisions, components, processes, risks, and quality requirements.

## 9. Architectural Decisions

- The system is initially implemented as a backend and API-centric solution.
- A frontend is part of a later expansion phase and not part of the initial roadmap.
- The solution shall intentionally be kept simple.
- Frameworks and external systems shall be used sparingly.
- Event Sourcing is considered as a business concept for traceability and consistency.
- Technological commitments to Event Sourcing mechanisms are intentionally avoided in this documentation.
- The architecture is consistently designed for testability.
- Unit tests are not optional but a binding part of the implementation.
- Full-text search is a mandatory component of the system.
- Automatic import from a monitored folder is a mandatory component of the system.
- Processed documents are transferred to a final, structured target storage.
- Final storage considers year-based folders and optionally import timestamps in filenames.
- Automation in development and deployment processes shall consider the existing Gitea/Actions environment.

## 10. Quality Requirements

Essential quality requirements are:

- Simplicity: The system shall be designed with minimal unnecessary complexity.
- Maintainability: Components shall be clearly separated and understandably structured.
- Testability: Every essential function shall be unit-testable.
- Traceability: Document changes and state transitions shall remain traceable through events.
- Consistency: States and transitions shall be robust and reconstructible.
- Automatability: Import, processing, testing, and deployment shall be automatable.
- Extensibility: A later frontend shall be able to build on the existing API and architecture.
- Searchability: Document contents shall be reliably discoverable via full-text search.
- Reliability: Automatic folder monitoring and processing shall function stably.

## 11. Risks and Technical Debt

Possible risks:
- Event Sourcing increases business and technical complexity if implemented too broadly or too early.
- OCR quality can vary depending on document quality.
- LLM-supported enrichment can yield varying results and must be understood in business context.
- Automatic file monitoring and relocation require robust error handling to avoid duplicates or inconsistent states.
- Without early testing, the risk increases that file processing, search indexing, and event flows become difficult to maintain.

Open items to be clarified later:
- Exact event model.
- Precise metadata model.
- Error handling strategy for partially failed processing.
- Detailed rules for the final folder structure.
- Search and ranking behavior of full-text search.

## 12. Glossary

- API: Application Programming Interface for using system functions.
- OCR: Optical Character Recognition; text recognition from documents or image contents.
- LLM: Large Language Model for semantic processing and enrichment of document contents.
- Event Sourcing: Business concept in which state changes are recorded as events.
- Full-Text Search: Search across the complete textual contents of documents.
- Import Folder: Monitored folder from which documents are automatically ingested.
- Year Folder: Target folder structure where processed documents are stored by year.
- Final Storage Structure: Final structured storage of processed documents.