B:\WORK.LOG

4 ENTRIES

B:\> log --graph

  • Developing and evaluating robot manipulation policies for a LeRobot SO-101 arm and a 7-axis Nero arm, trained on teleoperation demonstrations.
  • Led physical data collection: synchronized multi-camera episodes recorded with UMI.
  • Built and deployed a QC dashboard (FastAPI, SQLite, Vue 3 + TypeScript) that checks every uploaded episode through a versioned analysis registry: FFT audio cross-correlation for camera sync, OpenCV ArUco visibility, file integrity.

Novi Sad · robotics · imitation learning · fullstack

  • Co-authored a paper published at SQAMIA 2026 (CEUR-WS Vol-4336) on privacy-preserving food quality monitoring with federated learning and Fast-FedUL unlearning.
  • Built a PyTorch + Flower research framework that benchmarks exact retraining against approximate unlearning as clients join and leave, with sweeps over six model architectures.
  • Lead contributor to an async FastAPI platform: device telemetry and multi-tenant data across PostgreSQL, MongoDB and a FIWARE Orion-LD broker, JWT + Argon2 auth, Alembic migrations, per-company Grafana.
  • Food authenticity service: SIMCA over PCA on fused VIS, NIR and fluorescence spectra, with a versioned model registry.

Novi Sad · federated learning · unlearning · backend

B:\PAPER.TXT · the SQAMIA 2026 paper

(root) first commit: B.Sc. starts, 2021.09

B:\PAPER.TXTPUBLISHED Secure and Reversible FL forIoT-Driven Food Supply ChainsSQAMIA 2026 · CEUR Vol-4336OPEN A:\EDU.SYSMSC · BSC 2024 MSc Computer Science2021–2024 BSc Computer ScienceFaculty of Sciences, Novi Sad+ 9 coursesOPEN A:\SKILLS.ENVENV --WHERE LANG= Python C Java TypeScriptML= PyTorch Flower LeRobotGPU= WGPU/WGSL MPIWEB= DATA= OPS= ...OPEN