ASÀ propos de Aliaksandr
I train custom computer vision models on your data - detection, segmentation, classification, and OCR. Clear and fast delivery: baseline evaluation, measurable quality gains, and production-ready artifacts.
What I do:
1. Baseline in 48 hours: dataset audit, SOTA backbone (YOLO/RFDetr/SegFormer/ViT/SegNeXt), mAP/F1/IoU report, PR curves, FP/FN gallery.
2. Fine-tuning (upto 10 days): targeted augmentations, sampling/balance, LR scheduling, EMA/TTA; before/after comparison; trained checkpoint + inference.py.
3. Inference optimization: ONNX/TensorRT/OpenVINO, quantization, batch inference, CPU/GPU profiling.
4. MMDetection/MMSegmentation/YOLO/Roboflow training pipeline debugging: validation fixes, metric sanity checks, exports, strict reproducibility.
Tech Stack:
PyTorch, MMDetection, MMSegmentation, YOLO, RF-DETR, SegFormer, SegNeXt, ViT, OpenVINO, TensorRT, ONNX, Albumentations, OpenCV.
Deliverables
Structured report with metrics and error analysis, configs and scripts (train.py, eval.py, inference.py), pinned environments, trained checkpoints, exports (ONNX/TensorRT/OpenVINO), and a simple CLI for running inference.
Example Projects:
1. Overhead people detection for video/drone footage.
2. Deforestation change segmentation (SegNeXt/SegFormer) with rare-class F1 focus.
3. Receipt/invoice OCR (detector + CRNN/TrOCR fine-tune) with CER/WER tracking.
Process:
1. Dataset brief → 2) Baseline & error analysis → 3) Targeted fine-tuning → 4) Report (before/after, PR/ROC, FP/FN gallery) → 5) Artifacts & exports.
Guarantee:
I don’t promise a specific metric - I deliver transparent improvements over a baseline with fully reproducible training and inference.
Anglais
Bilingue ou natif
Russe
Bilingue ou natif
En télétravail uniquement
Travaille majoritairement à distance
Expériences
- Belarusian State UniversityPhD studentjanvier 2010 - novembre 2013 (3 ans et 10 mois)I have researched and implemented new adaptive wavelet function that increased vowels recognition rate
- L2labs OÜChief Executive Officerjuin 2022 - Aujourd'hui (4 ans)EstoniaCompany Business Development Manage and direct AI R&D team to create production ready neural network models and services based on SOTA Develop AI models for classification, semantic segmentation, object detection, GenAI and LLM Implement distributed AI model training framework for 2mln+ instances dataset Python/TF/PyTorch
- Fixate proSenior Deep Learning Engineermai 2021 - Aujourd'hui (5 ans et 1 mois)SpainImplemented Eye Tracking prediction neural network training on scale that increased final accuracy to 94%
Recommandations
Soyez le premier à recommander Aliaksandr
Contribuez à la réussite de ce freelance en partageant votre expérience de collaboration avec lui.
Ces profils de freelance correspondent également à vos critères
Agatha Frydrych
Backend Java Software Engineer
4.7
(3)
2
Baptiste Duhen
Fullstack developer
4.6
(4)
5
Amed Hamou
Senior Lead Developer
4
(2)
7
Audrey Champion
Web developer
4.3
(3)
4
Formations
- Master's degreeBelarusian State University2009Master's degree