Yesser Rizi
Data Science Engineer
C
Python
C++
SQL
Model Architecture & Training: 4/10
Active 1 day ago
+33 (7) 48423169 Invite to interview
Message
Download CVCV
Overview
Technical skills
Timeline
Roles
Overview
ML engineer at an early-to-mid career level specializing in building transformer-based chatbots and sequence models. The strongest proven skill is implementing and training a custom Transformer encoder-decoder with tokenization and a custom learning rate schedule, as shown in Plain-Transformers-based-Chatbot/Transformer.py. Public work lacks production-grade MLOps, experiment tracking, test coverage and advanced efficiency or security hardening evidence.
Technical skills
C
Python• Middle
C++• Junior
SQL• Junior
C++
PyTorch C++
TensorFlow C++
AI/ML
Anomaly Detection
Gymnasium
Stable-Baselines3
Streamlit
TF-Keras
NumPy
Pandas
LLM
Deep Learning
TensorFlow• 3y+
MLFlow
PyTorch
Scikit-learn
Transformers
Frontend
Frontend
JQuery
DevOps
AWS
Docker
Git
Analytics
Plotly
Power BI• 3y+
Timeline
R&D Engineer - Data Science, AI & Optimization
•
Middle
NAUTIA
•
Full-Time
Developed a reproducible Python analytics and experimentation pipeline for reinforcement learning sailboat agents, covering data generation through training, validation, evaluation, and visualization. Trained and validated a probabilistic machine learning model for predicting unstable operating conditions, using MLflow for tracking versions and metrics. Implemented anomaly detection and statistical validation methods, and integrated stability analysis and parameter-selection workflows into internal tools. Built an adaptive simulation algorithm that reduced training and evaluation time while preserving fidelity.
Python
MLFlow
Bordeaux I (jusqu'en 2014)
Master's Degree •
Applied Mathematics & Statistics - Image, Optimization and Data Science
Badji Mokhtar Annaba University
Master's Degree •
Artificial Intelligence & Information Processing
Research Project (Research & Innovation)
•
Middle
ReadRob
•
Full-Time
Collaborated with psychologists to define behavioral requirements and translate them into engagement indicators and evaluation criteria. Built and evaluated an LSTM-based recommendation system and developed a TensorFlow CNN for facial-emotion classification. Integrated hardware and software into a first prototype and created Power BI dashboards for costs and key performance indicators to support project monitoring.
TensorFlow
Power BI
Middle AI/ML Engineer
Confidence: Medium ML Engineer
ML engineer at an early-to-mid career level specializing in building transformer-based chatbots and sequence models. The strongest proven skill is implementing and training a custom Transformer encoder-decoder with tokenization and a custom learning rate schedule, as shown in Plain-Transformers-based-Chatbot/Transformer.py. Public work lacks production-grade MLOps, experiment tracking, test coverage and advanced efficiency or security hardening evidence.
Model Architecture & Training
4/10
How well models are designed and trained
Demonstrates hands-on model architecture and training work: a full Transformer encoder-decoder implementation, custom learning rate schedule, plus additional LSTM and MLP models in notebooks.
Evidence
Plain-Transformers-based-Chatbot/Transformer.py: class MultiHeadAttention, encoder/decoder, transformer()
Plain-Transformers-based-Chatbot/Transformer.py: class CustomSchedule and loss_function
Apparent-Temperature-forecasting/Apparent Temperature model/Code/apparent-temperature-forecasting.ipynb: temp_model LSTM training and fit()
Data Pipeline & Feature Engineering
3/10
How data is prepared for models
Solid data preprocessing and input pipeline work for NLP and tabular tasks using tokenization, tf.data, padding and sklearn scalers.
Evidence
Plain-Transformers-based-Chatbot/Transformer.py: textPreprocess, tokenize_and_filter, tf.data Dataset pipeline (cache, shuffle, batch, prefetch)
Apparent-Temperature-forecasting/Apparent Temperature model/Code/apparent-temperature-forecasting.ipynb: MinMaxScaler usage and feature creation
Experimentation & Evaluation
2/10
How results are measured and tested
Basic evaluation and plotting are present but there is no experiment tracking, ablation, or reproducible experiment management.
Evidence
Intrusion-Detection-Using-Deep-Learning-on-NSL-KDD-Dataset/NSL-KDD Model.ipynb: model.fit, model.evaluate, confusion_matrix
Apparent-Temperature-forecasting/Apparent Temperature model/Code/apparent-temperature-forecasting.ipynb: evaluation, predicted_values plotting
MLOps & Deployment
2/10
How models are shipped to production
Minimal deployment/ops artifacts: model save/load and a lightweight Streamlit interface, but no CI/CD, containerization, monitoring or versioning.
Evidence
Apparent-Temperature-forecasting/Apparent Temperature model/Code/apparent-temperature-forecasting.ipynb: model.to_json() and save_weights()
Plain-Transformers-based-Chatbot/StramlitBotInterface.py: Streamlit app wiring generate_response() to predict()
Computational Efficiency
3/10
How efficiently computing resources are used
Some attention to efficient training: distributed strategy awareness and tf.data performance primitives, but no profiling or advanced efficiency engineering.
Evidence
Plain-Transformers-based-Chatbot/Transformer.py: strategy = tf.distribute.get_strategy() and BATCH_SIZE = int(64 * strategy.num_replicas_in_sync)
Plain-Transformers-based-Chatbot/Transformer.py: dataset.cache(), dataset.batch(), dataset.prefetch(tf.data.experimental.AUTOTUNE)
Research Depth & Innovation
2/10
Depth of research and new ideas
Implements standard Transformer components correctly but follows common tutorial patterns without novel algorithmic contributions or published-research depth.
Evidence
Plain-Transformers-based-Chatbot/Transformer.py: end-to-end implementation of standard transformer building blocks (PositionalEncoding, MultiHeadAttention, encoder/decoder)
Plain-Transformers-based-Chatbot/Transformer.py: CustomSchedule which matches commonly used schedules in literature/tutorials
Expertise
LLM• Middle
Conversational AI & Chatbots• Middle
Industries
Artificial Intelligence• Middle
Cybersecurity• Middle
Technologies
Deep Learning
Python• Middle
SQL• Junior
C++• Junior
MLFlow
Scikit-learn
Transformers
TensorFlow• 3y+
Pandas
NumPy
Git
PyTorch
AWS
Docker
LLM
Stable-Baselines3
TF-Keras
TensorFlow C++
PyTorch C++
Streamlit
Anomaly Detection
Gymnasium
Recommendations
- Harden the training and evaluation pipeline with experiment tracking (W&B or MLflow), fixed random seeds, and reproducible dataset splits.
- Add unit and integration tests for core model components and preprocessing functions and include a small end-to-end CI job to validate training runs.
- Improve production readiness by containerizing the Streamlit service, adding model versioning and a simple inference API with latency tests and monitoring hooks.
- Profile and optimize training (GPU memory/profile traces, mixed precision, gradient accumulation) and add documented before/after performance numbers.
Repositories
The developer's experience in this domain has been verified based on AI analysis of the following repositories:
Junior Frontend Developer
Confidence: Medium UI Engineer
Frontend UI engineer (junior) focused on small browser-based interactive experiences with imperative DOM manipulation. The strongest proven skill is hands-on DOM and UI scripting as implemented in moneyBox/box.html through theMain and wins functions that build and animate the game UI. There is little evidence of testing, accessibility engineering, modern component architecture or build tooling in public code.
UI Component Architecture
2/10
How interface parts are built
Minimal componentization; DOM elements are created imperatively inside a single function without modular boundaries or reuse patterns.
Responsive & Cross-browser
1/10
Works on all screens and browsers
No responsive or cross-browser strategies; layout uses fixed physical units and absolute positioning rather than fluid or query-based approaches.
Performance Optimization
1/10
Speed of the interface
No measured performance or advanced optimizations; simple setInterval/setTimeout animations are used without cancellation or profiling.
Accessibility & Semantics
2/10
Usable for everyone
Small accessibility signal (tabIndex on a dice container) but otherwise missing ARIA, keyboard handlers, focus management and semantic roles.
State Management & Data Flow
1/10
Managing data in the app
State is maintained as local variables and DOM attributes; there is no structured state management, cancellation, or handling of async races.
UX & Visual Polish
3/10
Look and feel quality
Basic UX touches are present such as animated dice, slideDown and fadeIn transitions and visible score updates, but there are few edge states and no undo or perceived-performance considerations.
Industries
Gaming• Junior
Technologies
Frontend
JQuery
Recommendations
- Encapsulate UI pieces into modular components or functions and reduce global imperative logic so components can be unit tested and reused.
- Adopt semantic HTML and explicit keyboard/ARIA handling for interactive widgets, then run automated a11y checks (axe/CI) to improve accessibility.
- Migrate UI logic to a small modern framework or plain module-based code with clear state boundaries and introduce basic tests for game flows and edge cases.
- Replace fixed physical units with responsive layout techniques and add media queries or fluid sizing to improve cross-device support
Repositories
The developer's experience in this domain has been verified based on AI analysis of the following repositories: