How microblink/doc-ver-ops works
Project Type and Purpose
• This project isn’t a traditional full-stack “application” but rather a packaged deployment solution. Its primary purpose is to deploy a microservices-based document verification system using either Kubernetes (via Helm charts) or Docker Compose.
• The README clearly describes that you need a license key to enable the solution and that the system supports both scalable production environments (Helm with Kubernetes) and simpler docker-compose setups.
File Structure
• Notice the top-level directories and files such as “docker-compose” and “helm.” They indicate that the solution is delivered in a containerized form.
• Within the docker-compose directory, you see a base folder with several sub-folders for different service “bundles” (e.g., bundle-doc-ver, bundle-va, doc-ver-api, doc-ver-runner, embedding-store, retrieval-intermediary, visual-anomaly). Each bundle folder includes docker-compose.yaml files and service-specific subdirectories (like “nginx”, “proxy”, “model-serving”).
• The existence of configuration files (like .env files, nginx.conf, and proxy/bundle-signature.yaml) suggests that each component has a well-defined role and configuration file.
• In the Helm directory, the “doc-ver” chart shows templated Kubernetes configuration, indicating resource definitions, autoscaling setups, and secret management (e.g., for the licence key), which all support production-level deployment.
Main Components and Their Interactions
• Document Verification API (doc-ver-api)
• Document Runner (doc-ver-runner)
• Visual Anomaly detection (visual-anomaly)
• Embedding Store (embedding-store)
• Retrieval Intermediary (retrieval-intermediary)
• Model Serving components: This includes further subdivisions because both “bundle-doc-ver” and “bundle-va” directories have a “model-serving” folder that includes:
• Model download modules (script: model-download.sh, configuration: rclone.conf)
• Tensorflow-based model serving (with monitoring.txt to help observe performance)
• Proxy and NGINX Components: Each bundle also has nginx and proxy configuration – these indicate reverse proxy/load balancer functionalities.
• Incoming API traffic being routed (likely via an ingress controller or NGINX) towards the appropriate backend service.
• Inter-service communications between the retrieval intermediary, embedding store, and document verification API.
• Data flows from model serving (where models are downloaded, loaded, and served using TensorFlow) interacting with the document verification logic.
• The proxy components (with bundle-signature configurations) likely secure or modify the communication between services.
Architectural Patterns and Technologies
• Microservices architecture. Although every container is managed separately through docker-compose files or Kubernetes charts, the service decomposition follows clear boundaries.
• In a Helm deployment, include components such as autoscaling, resource limits, and possibly service meshes or ingress controllers (even if they’re not in the repository, they are inherently part of Kubernetes deployments).
• Technologies:
• Docker containers for packaging.
• Kubernetes for Helm-based deployments.
• NGINX as a reverse proxy/front end.
• TensorFlow for model serving.
• Scripts (like model-download.sh) for auxiliary functions like model retrieval.
Detailed Review and Final Touches
• Revisit each folder from the docker-compose and helm directories to ensure you capture all the individual services (for instance, distinguishing between bundle-doc-ver and bundle-va which may represent different verification modalities).
• To include auxiliary tools like the model download scripts, which are essential for the dynamic retrieval of ML models.