Documents

End‑to‑End OCR Domain Design & Implementation

Led the full design and implementation of the OCR domain inside BSDFlow – a large‑scale, microservices‑based management platform built in collaboration with Kaltura and KamaTech.

I designed a production‑ready OCR microservice in NestJS (TypeScript) that processes documents end‑to‑end, from upload to extracted structured text.

The service supports multiple file types including scanned PDFs, image files (JPG/PNG), and Office documents, and integrates Tesseract OCR and LibreOffice for document conversion and text extraction.

I focused on clean architecture, separation of concerns, and extensibility, so that new OCR providers or file pipelines can be added with minimal changes.

Smart “Quality Gate” Decision Engine (Local vs Cloud OCR)

Designed and implemented a smart quality‑gate pipeline that automatically decides whether to use embedded text, local OCR, or delegate to a cloud OCR provider.

The pipeline measures text coverage, confidence, and extraction quality, and applies clear heuristics to decide when the result is good enough and when a fallback is required.

This reduced unnecessary cloud calls, improved accuracy, and created a resilient, cost‑aware OCR flow.

The design uses Strategy and Adapter patterns so that each OCR provider and file type is handled through a clean, testable abstraction.

Integration with API Gateway, Caching & Observability

Integrated the OCR and document flows behind a central API Gateway, including request routing, health checks, structured error handling, and consistent API contracts for the frontend.

Implemented Redis‑based caching using file fingerprints (SHA‑256 hashes) to avoid reprocessing identical documents, significantly improving performance under load.

Added observability hooks with Prometheus metrics and structured logging, enabling monitoring of OCR paths, fallback rates, and system health for debugging and production readiness.

Testing, Docker & Reproducible Development Environment

Backed the service with Jest unit and integration tests, covering the OCR pipeline, quality gate, and provider adapters.

Built a full Docker & Docker Compose environment so the entire stack (OCR, Gateway, Redis, PostgreSQL, Kafka, monitoring) can be spun up locally in a consistent and reproducible way.

This enabled fast onboarding, reliable demos, and realistic production‑like testing for the whole team.

• Wrote Jest unit and integration tests and containerized the services with Docker and Docker Compose to support a consistent, team-wide development environment.

Backend development in C# / .NET as part of a Documents team

Developed a backend service communicating with a Worker that sends CUD commands

Designed and implemented data models using Entity Framework Code First

Created database tables to store and manage document-related data

Implemented a validation mechanism in the service layer

Extensive use of interfaces, abstraction, and polymorphism

Worked with a clean, layered architecture and maintainable code

Designed and implemented backend services in NestJS and C# (.NET) to manage the full document lifecycle: upload, storage, retrieval and deletion.

Integrated MinIO object storage with PostgreSQL for document metadata, permissions and controlled access to documents.

Developed an internal React-based UI for teams to browse, search, filter and manage documents.

Built event-driven workflows with Kafka to publish document events (uploaded / updated / deleted) and trigger asynchronous processing.

Collaborated in an Agile team environment, including code reviews, Git-based workflow and close mentorship from senior engineers.