Shakhansho Shukur Sabzaliev
AI Engineer
5+ years exp
SQL
Bash
Python
Data Pipeline & Feature Engineering: 5/10
Experimentation & Evaluation: 4/10
Active 16 days ago
+996 (555) 625835 Invite to interview
Message
Download CVCV
Overview
Technical skills
Timeline
Roles
Overview
A data-focused practitioner at a middle level who produces well-structured Kaggle-style notebooks and competitive-programming scripts. The strongest proven skill is tabular data work and feature engineering, evidenced by the Titanic and House Prices notebooks with custom transformers, imputation, and GridSearchCV-driven model selection. There is limited evidence of production deployment, experiment tracking, custom model architectures or GPU/efficiency engineering in public code.
Technical skills
SQL
Bash
Python• Middle • 5y+
Python
FastAPI
Databases
PostgreSQL
AI/ML
Computer Vision
LangChain
MediaPipe
NumPy
PyTorch
Scikit-learn
Streamlit
Synthetic Data
TensorFlow
YOLO
Pandas
Jupyter Notebook
Pillow
scikit-image
OpenCV
XGBoost
DevOps
Git• 5y+
Analytics
Matplotlib
Seaborn
QA
Pytest• 5y+
Timeline
ML Engineer
•
Middle
Business Boom
•
Full-Time
Architected end-to-end computer vision pipelines for client projects, covering dataset preparation, model training, and deployment of detection and classification solutions. Applied research-driven approaches to improve model accuracy through targeted data annotation and iterative experimentation. Partnered with cross-functional teams to prototype AI solutions aligned with business use cases.
Junior ML Engineer
•
Junior
Medgital
•
Full-Time
Developed a brain segmentation application using FastSurfer/FreeSurfer and integrated it for medical deployment with C#. Researched approaches for whole-body blood vessel segmentation and worked on curating medical imaging datasets to support training and evaluation.
C#
AI/ML Intern
•
Junior
BallerTV
•
Internship
Evaluated and calibrated pre-trained event detection models using volleyball footage to improve accuracy for sports broadcasting. Researched deep learning methods for image and video enhancement such as restoration, denoising, and super-resolution. Built AWS skills using core services to support ML workflows and storage needs.
Amazon EC2
Amazon Redshift
DynamoDB
ML Intern
•
Junior
ArtLabs
•
Internship
Co-developed an open-source Python data anonymization library, handling core tabular, image, and document anonymization features. Maintained the GitHub repository by managing issues and pull requests and supporting automated testing with pytest. Also worked on a computer-vision tennis player tracking system.
Python
Git
Pytest
Middle AI/ML Engineer
Confidence: Medium Data-centric
A data-focused practitioner at a middle level who produces well-structured Kaggle-style notebooks and competitive-programming scripts. The strongest proven skill is tabular data work and feature engineering, evidenced by the Titanic and House Prices notebooks with custom transformers, imputation, and GridSearchCV-driven model selection. There is limited evidence of production deployment, experiment tracking, custom model architectures or GPU/efficiency engineering in public code.
Model Architecture & Training
3/10
How well models are designed and trained
Practical model building with scikit-learn and XGBoost, ensembling and hyperparameter search are present, but no custom neural architectures or training loops are implemented.
Evidence
Kaggle/Titanic/notebook.ipynb: cross_val_score used across multiple models and model comparison blocks
Kaggle/Titanic/Titanic_pipeline.ipynb: Pipeline with CustomTransformer and LogisticRegression, GridSearchCV usage
Data Pipeline & Feature Engineering
5/10
How data is prepared for models
Clear, repeatable data preprocessing and feature engineering work - imputation, categorical encoding, custom transformers and domain-derived features are implemented.
Evidence
Kaggle/Titanic/notebook.ipynb: feature engineering (cabin_multiple, cabin_adv, numeric_ticket, name_title, norm_fare) and one-hot encoding pipeline
Kaggle/House Prices/dp.ipynb: extensive missing-value imputation, outlier removal, categorical handling and OneHotEncoder usage
Experimentation & Evaluation
4/10
How results are measured and tested
Structured experimentation with cross validation, GridSearchCV and ensembling plus simple performance reporting; no experiment tracking or reproducible run management is present.
Evidence
Kaggle/Titanic/notebook.ipynb: multiple GridSearchCV calls, clf_performance reporting function and cross_val_score outputs
Kaggle/Titanic/notebook.ipynb: construction and evaluation of several VotingClassifier ensembles
MLOps & Deployment
1/10
How models are shipped to production
Almost no deployment or MLOps artifacts - work is focused on notebooks and producing submission CSVs; no serving, versioning, CI/CD or monitoring code found.
Evidence
Kaggle/Titanic/notebook.ipynb: to_csv used to write submission files
Kaggle/Titanic/Titanic_pipeline.ipynb: Pipeline usage is present but no model serialization or serving code
Computational Efficiency
2/10
How efficiently computing resources are used
Minor efficiency considerations only - use of n_jobs and parallel GridSearchCV, but no GPU usage, profiling, quantization or memory-optimization work.
Evidence
Kaggle/Titanic/notebook.ipynb: GridSearchCV(..., n_jobs = -1) seen in multiple grid searches
Kaggle/Titanic/notebook.ipynb: use of cross_val_score and parallelized model evaluation
Research Depth & Innovation
1/10
Depth of research and new ideas
No signs of research-level innovation, paper re-implementations, or custom layers; the work follows standard textbook and Kaggle-style approaches.
Evidence
Kaggle/Titanic/Titanic_pipeline.ipynb: standard sklearn Pipeline and transformers (no custom architectures)
Kaggle/House Prices/EDA.ipynb: exploratory analyses and standard feature engineering, no paper-reproduction code
Verified artifacts
Expertise
MLOps & Model Lifecycle• Middle
Technologies
Python• Middle • 5y+
SQL
PostgreSQL
LangChain
OpenCV
YOLO
FastAPI
Computer Vision
TensorFlow
Git• 5y+
PyTorch
MediaPipe
Streamlit
Synthetic Data
Datasets• mentioned only
Recommendations
- Develop end-to-end production ML pipelines - add model serialization, CI, and a small serving component (FastAPI or BentoML) to bridge notebooks to production.
- Adopt experiment tracking (W&B or MLflow) and reproducible run configs to make hyperparameter searches and results auditable.
- Harden code by extracting reusable modules from notebooks into tested Python packages and add unit tests and basic CI.
- Continue strengthening feature engineering and cross-validation practice on tabular problems and learn workflow for model monitoring and drift detection.
Repositories
The developer's experience in this domain has been verified based on AI analysis of the following repositories:
Middle Data Scientist
Confidence: Medium ML Practitioner
A capable ML practitioner at an early-to-mid career level (middle tier) who builds end-to-end experiments and models inside notebooks. The strongest proven skill is applied predictive modeling and feature engineering, supported by the Titanic pipeline (CustomTransformer + sklearn Pipeline) and the Kaggle notebooks that show hyperparameter search and ensembling. There is limited evidence of productionization, rigorous uncertainty analysis, or domain-specific, production-grade system design in public code.
Statistical Rigor
4/10
Correct use of statistics
Shows basic statistical rigor: proper train/test splits, cross-validation and hyperparameter search, some distributional transforms (log). Lacks deeper uncertainty quantification, formal assumption checks, or causal analysis.
Data Wrangling & Cleaning
6/10
Preparing and cleaning data
Solid hands-on data wrangling: missing-value strategies, targeted imputations, explicit feature engineering for domain-specific fields (tickets, cabins, titles). Work is exploratory and notebook-centered rather than production ETL.
Evidence
shukkkur/Kaggle/House Prices/dp.ipynb: extensive missing-value handling (groupby median for LotFrontage, fillna strategies) and OneHotEncoder usage
shukkkur/Kaggle/Titanic/notebook.ipynb: feature engineering for cabin, ticket_letters, numeric_ticket, name_title
shukkkur/Kaggle/Titanic/Titanic_pipeline.ipynb: CustomTransformer class implementing feature creation
Exploratory Analysis & Visualization
6/10
Exploring and visualizing data
Exploratory analysis is purposeful with many visualizations and inline interpretations (heatmaps, pairplots, bar/boxplots) that connect observations to modeling decisions.
Predictive Modeling
6/10
Building models that predict
Predictive modeling discipline is evident: baseline models, many ML algorithms tried, GridSearchCV, ensembling and feature-importance inspection. Lacks deeper error analysis, calibration, or deployment-focused work.
Evidence
shukkkur/Kaggle/Titanic/notebook.ipynb: multiple models (LogisticRegression, RandomForest, SVC, XGB) with GridSearchCV and VotingClassifier ensembles
shukkkur/Kaggle/Titanic/Titanic_pipeline.ipynb: end-to-end sklearn Pipeline combining CustomTransformer, ColumnTransformer, and classifier
shukkkur/Predict-Species-from-Images/notebook.ipynb: HOG feature extraction, PCA, SVM training and ROC/AUC reporting
Business Insight & Impact
2/10
Turning analysis into business value
Little evidence of business-metric framing, cost-sensitive reasoning, or actionable recommendations tied to KPIs; work is framed as Kaggle/academic problems rather than product impact.
Reproducibility & Notebook Hygiene
3/10
Clean, repeatable analysis
Notebooks are runnable and include random_state seeds and pipelines, but there is no pinned environment, dependency management, CI, or productionization artifacts for reproducible deployment.
Technologies
Jupyter Notebook
XGBoost
Scikit-learn
Seaborn
Matplotlib
Pandas
NumPy
Pillow
scikit-image
Datasets• mentioned only
Recommendations
- Lead prototyping work for tabular and image ML models - feature engineering, CV, and model selection (use the Titanic and bees notebooks as starting patterns).
- Convert notebook pipelines into modular, testable Python packages (move CustomTransformer and preprocessing into reusable modules and add unit tests).
- Add reproducibility and deployment hygiene - pin dependencies, provide requirements/dockerfile, add scripts to run training reproducibly, and include model evaluation/validation reports.
- Work on error analysis and calibration tasks - per-class confusion analysis, calibration plots, and cost-sensitive metrics for production readiness.
Repositories
The developer's experience in this domain has been verified based on AI analysis of the following repositories: