Yasser El Haddar

Tech Lead - Machine Learning Ops Engineer

Transforming data into intelligence through cutting-edge machine learning, computer vision, and AI-driven solutions that push the boundaries of what's possible.

Download CV

About Me

terminal@yasser:~

█

About

Tech Lead at Mindtrace, building ML systems for IBM, NV5, and ESmart. Specialized in MLOps, computer vision, and production ML pipelines. Creator of ShopSense-AI, CacheFuse (60-90% cost reduction), SkewSentry, and MCP DS Toolkit. Writing about CV fundamentals on Medium. Manchester, UK.

Tech LeadershipMLOpsComputer VisionGCP/Cloud

Technical Skills

AI/ML

Computer VisionLLM DevelopmentPyTorch/TensorFlowModel Training

Engineering

MLOps & PipelinesFastAPI/MicroservicesPythonDocker/Kubernetes

Cloud & Infrastructure

GCP/Cloud ArchitectureServerlessCI/CDMonitoring

Experience

Technology Campus of Plattling

Research Assistant (Master Thesis)

2021-01 - 2021-06

Experimented with Generative Adversarial Networks (GANs) to enhance simulated LiDAR and RADAR point cloud data for realistic environmental representation. Transformed low-density sensor data into higher density formats to improve annotation and segmentation accuracy for autonomous driving applications.

GANs (Generative Adversarial Networks)KerasPyTorchCarla Simulator

Vitesco Technologies

Data Scientist

2021-06 - 2022-03

Developed ML-based approach for predicting defects in SCR Injector manufacturing, analyzing welding parameters (voltage, current, feed rate). Implemented and compared Random Forest, SVM, and XGBoost models. Utilized SPOT for dataset balancing and LIME for model explainability to optimize manufacturing quality control.

Random ForestSVMXGBoostSPOT+6

Mindtrace Limited

Tech Lead - Senior ML Ops Engineer

2022-03 - Present

Leading ML operations for asset inspection and manufacturing defect detection serving IBM, NV5, ESmart, and SharperShape. Architected serverless GCP pipelines, implemented YOLO/Dinov2 models achieving 92% accuracy, and enhanced point cloud segmentation (50% faster processing, 60% cost reduction). Built comprehensive MLOps infrastructure with Docker, Airflow, MLFlow, and FastAPI microservices.

PythonBash ScriptPyTorchGCP+15

Projects

ShopSense-AI

Microservices-based AI shopping platform with fine-tuned LLMs using QLoRA on Mistral-7B. Features semantic search with Qdrant vector database, real-time product discovery across multiple retailers (Amazon, Best Buy, Walmart), and intelligent recommendations combining vector similarity with LLM-generated advice. Achieves sub-50ms response times processing 100+ products per query.

PythonFastAPIDockerQdrant+10

CacheFuse

Production-ready LLM caching framework achieving 60-90% API cost reduction and 100x latency improvement (3ms vs 2-5s). Features intelligent cache invalidation with TTL management, multi-backend architecture (SQLite/Redis), privacy-compliant hash-only mode, and drop-in decorator system for zero-configuration setup.

PythonRedisSQLiteOpenAI API+4

SkewSentry

Automated ML feature validation system preventing training/serving skew in production deployments. Addresses 70% of ML failures through pre-deployment detection with multi-source adapter pattern, configurable tolerance thresholds, and comprehensive CI/CD integration generating HTML/JSON reports.

PythonFastAPIPydanticpandas+4

Football Smart Tactician

End-to-end ML-powered football analysis system using YOLOv11 for multi-class detection and pose estimation. Features advanced computer vision pipeline with homography projection, comprehensive analytics (heat maps, formation analysis, pass networks), OpenAI LLM integration for tactical commentary, and MLOps pipeline with Label Studio and MLFlow.

PythonFastAPIReactDocker+8

MCP DS Toolkit Server

Standalone Model Context Protocol server enabling AI assistants (Claude Desktop, Cursor) to perform complete data science workflows through natural language. Features 29 specialized tools for data loading, preprocessing, model training (14+ algorithms), and experiment tracking with SQLite-based persistence.

PythonMCP ProtocolSQLitescikit-learn+4

Personal AI Newsletter

AI-powered newsletter generation system that automatically curates personalized content from GitHub, Reddit, Hacker News, and web articles. Uses GPT-4 for intelligent scoring, LangGraph for workflow orchestration, and delivers 20 curated articles in 45-60 seconds with built-in scheduling and analytics.

PythonLangGraphOpenAIMCP+4

Finance Web App

Streamlit web application for real-time stock visualization and prediction. Features RAG-enhanced chatbot powered by GPT-3.5 Turbo for EDGAR filing queries, daily stock market forecasts using Temporal Fusion Transformer, and real-time monitoring of intraday prices, forex rates, and market news.

PythonStreamlitLangChainOpenAI API+5

Mindtrace

Enterprise ML platform for asset inspection and manufacturing defect detection serving IBM, NV5, ESmart, and SharperShape. Features point cloud segmentation (50% faster), 3D reconstruction with COLMAP, serverless GCP pipelines, and comprehensive MLOps infrastructure.

PythonBash ScriptPyTorchGCP+15

Tech Stack

AI/ML & Deep Learning

PyTorchTensorFlowKerasscikit-learnComputer VisionLLM DevelopmentGANsYOLOv11Dinov2Model TrainingFine-tuningQLoRAMistral-7BMLFlowWandBLabel StudioOpenCVPDALHydraCOLMAPHLoc

Data Science

PandasNumPyMatplotlibPyTorch ForecastingRandom ForestSVMXGBoostSPOTLIME

Backend Engineering

PythonFastAPIMicroservicesBash Scriptasync/awaitpytestPydanticPyYAMLCeleryOAuth2RESTful APIsSQLAlchemyAlembic

Cloud & DevOps

GCPServerlessCI/CDGitHub ActionsAirflowDockerKubernetesNginxUVuvx/uvMCP ProtocolMLOps

Databases & Caching

PostgreSQLSQLiteQdrantRedisVector DBs

AI APIs & LLM Frameworks

OpenAI APIAnthropic APIHugging Face Transformerssentence-transformersAlpha Vantage APIHuggingface datasetLangGraphLangChainRAG SystemsResendFirecrawl

Web & Frontend

ReactNext.jsTypeScriptStreamlit

Click on any skill to see how I've used it

Knowledge Fragments

Deep LearningComputer Vision

AlexNet: The Deep Learning Revolution That Sparked The Boom

In the annals of artificial intelligence, certain moments stand out as pivotal turning points. The year 2012 marked one such moment, thanks to AlexNet - the breakthrough that ignited the deep learning revolution and changed computer vision forever.

Nov 8, 2025

CNNComputer Vision

LeNet-5: The Pioneering CNN That Started It All

In the vast and ever-evolving landscape of artificial intelligence, certain landmarks stand out as pivotal moments of innovation. LeNet-5 was the pioneering convolutional neural network that laid the foundation for modern deep learning architectures.

Nov 5, 2025

Object DetectionHOG

HOG: The Foundation of Early Object Detection

Object detection is everywhere now. From identifying pedestrians in self-driving cars to tagging faces on social media, the technology has become ubiquitous. But it all started with Histogram of Oriented Gradients (HOG) - a breakthrough that shaped modern computer vision.

Nov 4, 2025

AI FrameworksMultimodal Learning

4M: The AI Framework Revolutionising Multimodal Learning

In the rapidly evolving landscape of artificial intelligence, the ability to seamlessly integrate and understand multiple data modalities is becoming increasingly crucial. Enter 4M - a revolutionary framework that is reshaping how we approach multimodal AI.

Jan 8, 2025

PythonDevOps

uv: Redefining Python Environment and Dependency Management

In Python development, managing dependencies and environments is crucial to ensure project isolation and reproducibility. Virtual environments and tools like pip have been standard, but uv is revolutionizing how we approach Python package management.

Dec 26, 2024

SIFTFeature Detection

SIFT: Feature Detection That Revolutionized Image Matching

Imagine a world where computers can effortlessly identify objects in images, regardless of changes in scale, rotation, or even lighting conditions. This is the reality made possible by SIFT (Scale-Invariant Feature Transform), a groundbreaking algorithm that revolutionized computer vision.

Nov 1, 2024

Edge DetectionCanny Algorithm

Edge Detection & Canny Algorithm: How We First Taught Computers to See Edges

Imagine a world where computers can only see a jumble of pixels. No shapes, no objects, just a sea of numbers representing color values. Edge detection, particularly the Canny algorithm, was the breakthrough that taught computers to see boundaries and shapes.

Oct 27, 2024

Writing & Insights

Dive into my technical articles covering machine learning, AI engineering, and production systems. Each card represents a knowledge fragment from my journey on tech.

Articles

Topics

The Portal

Let's Connect

Your message will travel through this portal to reach me

Send a Message

Ready to discuss your next project or just say hello? Drop me a line!