Roman Melnikov
builds the backend.

A Django platform that turns raw punches from biometric terminals into schedules, timesheets and payroll-grade reports for distributed organisations.

• Device layer: a long-running iClock/ZKTeco protocol handler ingesting terminal events over unreliable links, with idempotent replay and per-device state.
• Rules engine: shifts, rotations, overtime, absences and exceptions computed per employee across timezones.
• Reports: a unified report layer (timesheet, extended, attendance) exporting to Excel with consistent totals.
• Async core: Celery + Redis for imports, recalculation and scheduled jobs; Channels for live terminal status.

Automated preparation of employee biometrics: detect, validate and crop faces to the exact geometry terminals accept, so enrolment stops being a manual Photoshop job.

• Detection: MTCNN + OpenCV pipeline with quality gates (size, angle, sharpness, single-face check).
• Delivery: normalised images pushed to devices and archived to S3-compatible storage.

Glue that removes humans from routine loops: partner APIs, catalogue synchronisation, scheduled data exchange and a Telegram bot for alerts and on-call reporting.

• Reliability: retries with backoff, dead-letter handling, structured logs and health endpoints.
• Scheduling: crontab-driven periodic tasks managed from the admin instead of the server.