Essohanam Josué KPATCHA

Profile

Final-year Data Engineering student at ENSA Al Hoceima with a consistent track record of translating complex datasets into actionable business recommendations. Hands-on experience in analytics pipelines, predictive modeling, and executive reporting — with direct exposure to BCG consulting frameworks. Strong analytical mindset, structured communication, and results orientation. Available immediately · Casablanca CFC.

Technical Skills

Analytics & ML

Python (Pandas, NumPy, Scikit-learn, XGBoost, LightGBM), SQL (advanced), Statistical Modeling, Feature Engineering, A/B Testing, Classification, Clustering, Regression

Data Visualization & BI

Power BI, Apache Superset, Plotly, Matplotlib, Seaborn — interactive dashboards, KPI reporting, executive storytelling

Data Engineering

ETL/ELT pipelines (Airflow, dbt, Airbyte), Apache Kafka, Apache Spark, Azure Databricks, Snowflake, PostgreSQL, Great Expectations

AI & Generative AI

LangChain, LangGraph, RAG, LLMs (Groq, Gemini), Langfuse, ChromaDB, HuggingFace, MCP

Cloud & DevOps

GCP, Azure (Databricks), AWS, Docker, Kubernetes, Terraform, CI/CD

Professional Experience

AI & Data Engineering Intern

Jun – Sep 2025 · 4 months

ISICOD — Rabat, Morocco

Co-built an end-to-end biometric AI platform for fingerprint classification (12 classes) and ridge counting — processing 15,848 images from 5 heterogeneous sources. Produced the largest labeled dataset of its scope in the literature.
Delivered a production REST API (FastAPI + React) with real-time inference, model performance dashboard, and admin monitoring — directly operational for the product team.
Sole implementer of a Human-in-the-Loop retraining system: user corrections trigger incremental PyTorch retraining, enabling continuous model improvement without static dataset dependency.

Stack: Python · PyTorch · FastAPI · React · PostgreSQL · Docker

Machine Learning Engineer Intern

Jun – Aug 2024 · 2 months

3D Smart Factory — Mohammedia, Morocco

Designed custom Kolmogorov–Arnold Network (KAN) layers integrated into a 3D U-Net for volumetric medical image segmentation — achieving +3% accuracy gain over the standard 3D U-Net baseline.
Optimized architecture through systematic tuning of depth, kernel size, and activation strategies using PyTorch on MRI datasets.

Stack: Python · PyTorch · 3D U-Net · KAN · Medical Imaging · NumPy

Selected Projects

PowerCo Churn Analysis — BCG Data Science Simulation

Mar 2026

BCG X Analytics Framework · Consulting Simulation

Reframed client's hypothesis: diagnosed churn drivers across 14,600+ customers and 193,000 pricing records — found that margin and segment, not price, are the primary drivers.
Engineered 50+ features, benchmarked RF/GB/XGBoost — selected XGBoost: ROC-AUC = 0.68, Precision = 0.96. Delivered executive summary with targeted retention recommendation.

Stack: Python · XGBoost · Pandas · Matplotlib · Seaborn

AI Financial Analyzer — BCG GenAI Consulting Simulation

Feb 2026

BCG GenAI Team · SEC EDGAR · RAG Architecture

Built a production-grade financial intelligence system answering the core consulting question "How is this company performing?" — extracting 2,000+ records from 34 companies × 3 years via XBRL mapping.
Engineered 20+ financial KPIs (ROE, ROA, net margin, CAGR) benchmarked by industry. Deployed hybrid RAG engine (ChromaDB + SQLite, Groq Llama 3.3-70b) with Langfuse observability and LiteLLM failover.

Stack: Python · LangChain · ChromaDB · SQLite · Groq · Langfuse · Flask · React

Financial Health Prediction — Zindi Africa Competition

Jan 2026

SME Credit Risk · Southern Africa · 9,600 businesses

Built a predictive classification system for SME financial health across 4 countries (9,618 records, 47 engineered features). Gradient Boosting selected: F1 = 0.82 weighted, 78.7% accuracy.
Deployed end-to-end FastAPI scoring API with batch prediction endpoints and 1-page executive dashboard for stakeholder alignment.

Stack: Python · XGBoost · scikit-learn · FastAPI · Pandas

Togo National Internet Connectivity — Demand Prediction

May–Jun 2024

Zindi Africa Competition · National Policy · FTTH Infrastructure

Built ML pipeline on national survey dataset (~14,400 rows, ~4,000 features) to predict household internet uptake guiding government FTTH investment decisions. XGBoost: ROC-AUC = 0.86, competition score 0.7144.

Stack: Python · XGBoost · PCA · Pandas · Scikit-learn

Education

Ecole Nationale des Sciences Appliquées d'Al Hoceima (ENSA)

Engineer's Degree — Data Engineering

Oct 2023 – Jun 2026

Ecole Nationale des Sciences Appliquées de Fès (ENSA)

Integrated Preparatory Classes — Applied Mathematics

Oct 2021 – Jun 2023

Certifications

Oracle Cloud Infrastructure 2025 — Certified Generative AI Professional · Spark, Hadoop & Snowflake for Data Engineering — Duke University · AI Agents Fundamentals — HuggingFace · AWS Cloud Quest: Cloud Practitioner — Amazon Web Services · Neo4j Certified Professional · ETL and Data Pipelines with Shell, Airflow & Kafka — IBM · Introduction & Intermediate Machine Learning — Kaggle / Zindi Africa

ESSOHANAM JOSUÉ KPATCHA