Gloryson Horace

AI/ML • Middle

5+ years exp

10+ projects

Rust

Python

Data Pipeline & Feature Engineering: 5/10

Model Architecture & Training: 4/10

Experimentation & Evaluation: 4/10

Active 1 day ago

Invite to interview

Message

Download CVCV

Overview

Technical skills

Roles

Overview

ML Engineer (approx mid-level) focused on practical PyTorch training pipelines and hands-on deep learning education. The strongest proven skill is building end-to-end training and evaluation pipelines with the StepByStep trainer, TensorBoard integration and checkpointing as implemented in stepbystep/v0.py and the chapter notebooks. There is limited evidence of production-grade deployment, distributed training, end-to-end experiment management, or original research contributions in the public code.

Technical skills

Rust• Junior

Python• Senior • 5y+ • 10+ projects

Rust

Cargo

Databases

PostgreSQL• 5+ projects

AI/ML

Deep Learning

NumPy

Scikit-learn

Torchvision

CUDA

Pandas

Pillow

Computer Vision• 3y+ • 5+ projects

NLP• 3y+ • 5+ projects

DevOps

Kubernetes• 3y+ • 4 projects

Analytics

Matplotlib

Middle AI/ML Engineer Confidence: High ML Engineer

Model Architecture & Training

4/10

How well models are designed and trained

Solid hands-on model training and architecture practice for small models: custom training loops, loss/optimizer choices and wrapper utilities are present, but no novel architectures or large-scale training design.

Evidence

pytorch-step-by-step-learning/stepbystep/v0.py: StepByStep class with _make_train_step_fn, train, save_checkpoint, load_checkpoint

deep-learning-specialization-coursera/nn_and_dl/logistic_regression.ipynb: custom sgd implementation and sigmoid/loss derivation

pytorch-step-by-step-learning/chapters/chapter_2/chapter_2_colab.ipynb: make_train_step_fn and multiple model_configuration/*.py training variants

Data Pipeline & Feature Engineering

5/10

How data is prepared for models

Practical data preparation and augmentation pipelines for small datasets including custom dataset classes, transforms, samplers and balanced sampling.

Evidence

pytorch-step-by-step-learning/chapters/chapter_4/chapter_4.ipynb: TransformedTensorDataset and Compose transforms pipeline

pytorch-step-by-step-learning/data_generation/simple_linear_regression.py: synthetic data generation and train/val split

pytorch-step-by-step-learning/chapters/chapter_4/chapter_4.ipynb: WeightedRandomSampler class imbalance handling and sampler usage in DataLoader

Experimentation & Evaluation

4/10

How results are measured and tested

Basic experiment tracking and evaluation are implemented with TensorBoard logging, plotting utilities and evaluation examples, but no experiment management system or reproducible experiment registry is present.

Evidence

pytorch-step-by-step-learning/stepbystep/v0.py: set_tensorboard, writer.add_scalars, add_graph

pytorch-step-by-step-learning/chapters/chapter_3/chapter_3.ipynb: confusion_matrix and ROC/precision-recall evaluation usage

pytorch-step-by-step-learning/chapters/chapter_2/chapter_2_colab.ipynb: writer usage and runs/simple_linear_regression TensorBoard example

MLOps & Deployment

3/10

How models are shipped to production

Basic MLOps practices: checkpointing, save/load of model and optimizer states, and a predict wrapper exist, but there is no serving, model versioning, CI/CD or drift monitoring.

Evidence

pytorch-step-by-step-learning/stepbystep/v0.py: save_checkpoint and load_checkpoint implementations

pytorch-step-by-step-learning/chapters/chapter_2/chapter_2_colab.ipynb: torch.save(model_checkpoint.pth) and restore flow

Computational Efficiency

3/10

How efficiently computing resources are used

Awareness of device placement and per-batch GPU usage is present and there are notes about not loading entire datasets to GPU, but there is little evidence of profiling, quantization experiments with measured gains, or distributed training.

Evidence

pytorch-step-by-step-learning/chapters/chapter_2/chapter_2_colab.ipynb: device detection, moving batches to device and explicit comments on not loading entire dataset to GPU

pytorch-step-by-step-learning/chapters/chapter_2/chapter_2_colab.ipynb: use of torch.cuda.is_available and examples showing CUDA tensors

Research Depth & Innovation

1/10

Depth of research and new ideas

Primarily tutorial and instructional material; there is little research depth or original algorithmic innovation beyond faithful implementations and didactic examples.

Evidence

deep-learning-specialization-coursera/nn_and_dl/logistic_regression.ipynb: derivation and didactic SGD implementation

pytorch-step-by-step-learning/chapters/*: educational notebooks implementing standard CNN/ML building blocks

Expertise

Computer Vision & Image Analysis• Middle

MLOps & Model Lifecycle• Middle

Recommendations

Develop small-to-medium computer vision models and training pipelines (data generation, augmentation, balanced sampling, training loops, TensorBoard) for prototyping and experiments.
Build reusable model-training libraries and unit-tested utilities (extend StepByStep into a well-documented package with tests and CI).
Implement lightweight MLOps workflows: model versioning, reproducible experiment runs (W&B or MLflow), and a simple serving endpoint for model inference.
Work on measurable efficiency improvements: add profiling, batch optimizations, and basic quantization experiments with before/after metrics.

Repositories

The developer's experience in this domain has been verified based on AI analysis of the following repositories:

Gson-glitch deep-learning-specialization-coursera pytorch-step-by-step-learning

Middle Data Scientist Confidence: Medium ML Practitioner

A middle-level deep learning practitioner focusing on hands-on PyTorch training workflows and model experiments. The strongest proven skill is building and iterating custom training and experiment workflows, exemplified by the StepByStep training class and the modular training scripts. Missing from public work is productionization evidence - distributed training, MLOps pipelines, dependency pinning, automated testing and formal statistical uncertainty analysis are not shown.

Statistical Rigor

2/10

Correct use of statistics

Basic statistical treatment and evaluation are present with attention to numerical stability, but there is no formal uncertainty quantification, hypothesis testing or causal analysis.

Evidence

deep-learning-specialization-coursera/nn_and_dl/logistic_regression.ipynb: analytic derivation of loss/gradient and numerical examples

pytorch-step-by-step-learning/chapters/chapter_3/chapter_3.ipynb: confusion_matrix, ROC and precision-recall calculations

Data Wrangling & Cleaning

6/10

Preparing and cleaning data

Good data-handling practices for experiments: deterministic seeding, explicit train/validation splits, transforms, handling class imbalance, and conscious GPU/CPU data movement.

Evidence

pytorch-step-by-step-learning/stepbystep/v0.py: set_seed and device handling

pytorch-step-by-step-learning/chapters/chapter_4/chapter_4.ipynb: TransformedTensorDataset, data transforms and WeightedRandomSampler

pytorch-step-by-step-learning/data_preparation/v2.py: train/val split and DataLoader construction

Exploratory Analysis & Visualization

5/10

Exploring and visualizing data

Exploratory plots and visualizations are frequent and accompanied by explanatory text; they are informative but mainly tutorial-style rather than targeted data storytelling for stakeholders.

Evidence

deep-learning-specialization-coursera/nn_and_dl/logistic_regression.ipynb: scatter plots and decision boundary visualization

pytorch-step-by-step-learning/chapters/chapter_4/chapter_4.ipynb: image grids and transformation visualizations

pytorch-step-by-step-learning/chapters/chapter_3/chapter_3.ipynb: make_moons scatter and confusion matrix heatmap

Predictive Modeling

5/10

Building models that predict

Strong hands-on predictive-model workflow skills: baseline-to-deep progression, custom training loops, checkpointing, TensorBoard logging and evaluation; missing advanced model tuning, rigorous CV pipelines or production deployment evidence.

Evidence

pytorch-step-by-step-learning/stepbystep/v0.py: StepByStep training loop, checkpoint save/load, predict and plot_losses

pytorch-step-by-step-learning/chapters/chapter_4/chapter_4.ipynb: logistic and deeper models using BCE/BCEWithLogits loss

pytorch-step-by-step-learning/model_training/v2.py and model_training/v5.py: incremental training scripts with mini-batching and TensorBoard logging

Business Insight & Impact

1/10

Turning analysis into business value

Minimal business framing; problems are toy-focused and there is no discussion of product metrics, cost of errors, or business impact.

Evidence

pytorch-step-by-step-learning/chapters/chapter_4/chapter_4.ipynb: explicit toy problem statement 'Is the line diagonal?'

Reproducibility & Notebook Hygiene

4/10

Clean, repeatable analysis

Notebooks and scripts include reproducibility measures like seeds, checkpointing, modularized writefiles and TensorBoard, but pinned environments, dependency manifests and automated tests are missing.

Evidence

pytorch-step-by-step-learning/stepbystep/v0.py: set_seed and save_checkpoint/load_checkpoint methods

pytorch-step-by-step-learning/chapters/chapter_2/chapter_2_colab.ipynb: use of %%writefile to create modular scripts for reproducible runs

pytorch-step-by-step-learning/model_training/v5.py: TensorBoard writer usage and checkpoint saving

Expertise

Analytics• Middle

Technologies

Deep Learning

Scikit-learn

Matplotlib

Pandas

NumPy

Pillow

Torchvision

CUDA

PyTorch• mentioned only

Recommendations

Harden a training pipeline for production: add requirements.txt or pyproject, Dockerfile, simple CI and an experiment-tracking backend (MLflow or Weights & Biases).
Add reproducible hyperparameter tuning and robust CV (or time-series splits when relevant) plus structured error analysis and calibration reports.
Implement unit tests for core training utilities and add a pinned environment to improve reproducibility for collaborators and hiring evaluations.
Work on one end-to-end project using a real dataset and include business-metric framing, FP/FN cost analysis and deployment notes to demonstrate product impact.

Repositories

The developer's experience in this domain has been verified based on AI analysis of the following repositories:

Gson-glitch deep-learning-specialization-coursera pytorch-step-by-step-learning

Junior Backend Developer Confidence: Medium Generalist

Algorithm-focused developer at a junior level with a clear strength in coding-challenge problem solving and Rust learning. The strongest proven skill is algorithm design and implementation as demonstrated by numerous Python solutions such as threeSum and fourSum. There is no evidence of production backend work like APIs, databases, deployments, observability or resilience patterns in public code.

API Design

How well APIs are designed

Not evidenced in public code

Data Layer & Database

Working with databases

Not evidenced in public code

Scalability & Performance

2/10

Handling load and speed

Evidence of algorithmic complexity awareness and micro-optimizations in problem solutions, but no production-grade scalability patterns like caching, queues or load testing.

Evidence

Gson-glitch/neetcode-submissions/Data Structures & Algorithms/three-integer-sum/submission-5.py: two-pointer threeSum implementation that sorts input and skips duplicates Gson-glitch/neetcode-submissions/Data Structures & Algorithms/4sum/submission-0.py: sorted iteration with count bookkeeping to avoid extra nested work Gson-glitch/rust_learning/src/bin/hash_map.rs: notes on SipHash vs FxHash and performance trade-offs for hashing

System Architecture

Overall system structure

Not evidenced in public code

Security & Auth

Protecting data and access

Not evidenced in public code

Reliability & Observability

Stability and monitoring

Not evidenced in public code

Expertise

Python• Junior

Rust• Junior

Technologies

Rust• Junior

Cargo

Recommendations

Develop algorithmic libraries, coding-challenge resources, or interview-prep tooling in Python to leverage proven strengths.
Build small-to-medium Rust CLI tools or utility libraries that consolidate learning exercises into reusable modules and unit tests.
Work on end-to-end small services (simple REST APIs or background workers) to gain experience with persistence, observability and production resilience patterns.

Repositories

The developer's experience in this domain has been verified based on AI analysis of the following repositories:

Gson-glitch neetcode-submissions rust_learning