// AI ENGINEER — MILAN, ITALY

Fauzan
Ejaz

Turning raw signals into forecasts, clusters and decisions— machine learning & MLOps for the real world.

view_projects( ) →github

scroll to explore the data ↓

+0%

RMSE / R² lift via cluster-based forecasting

F1-score — BERT sentiment on 360K+ reviews

mAP — YOLOv8 visual quality control

0K+

Amazon reviews processed with NLP

[01]

About

Data-driven professional with hands-on experience in data analysis, statistical modeling, and machine learning. Transformed into a Data Scientist. Skilled in Python, SQL, and predictive analytics with a strong foundation in automation and MLOps practices. Passionate about transforming complex data into actionable insights that enhance decision-making, optimize performance, and drive business growth within collaborative, innovation-focused teams.

📍 MILAN, Italy+39 3421014345+91 9832887252

lang_proficiency.plot( )

ItalianB1

EnglishC1

ArabicA1

HindiC1

UrduC1

BengaliB1

[02]

Skills

stack.stream( ) — hover to pause

PythonRSQLScalaApache SparkETL PipelinesData WarehousingAWS (S3, EC2, Lambda, Redshift)Machine LearningGenerative AIPythonRSQLScalaApache SparkETL PipelinesData WarehousingAWS (S3, EC2, Lambda, Redshift)Machine LearningGenerative AI

Big Data ProcessingPandasNumPyMatplotlibPower BIModel OptimizationCI/CDGitAgile & ScrumBig Data ProcessingPandasNumPyMatplotlibPower BIModel OptimizationCI/CDGitAgile & Scrum

soft_skills[ ]

Analytical ThinkingProblem SolvingOwnership & AccountabilityCollaboration & CommunicationTime ManagementAdaptabilitySelf-Motivated & Goal-Oriented

[03]

Experience

JULY 2025 – JUNE 2026 · NAPLES, ITALY

AI/ML Engineer

@ GEKO S.p.A — Energy & Ambient

▸Developed and deployed electricity-consumption forecasting models (XGBoost, Scikit-learn) with Optuna hyperparameter optimisation for seasonal Italian energy-demand patterns, Built an automated model evaluation and monitoring pipeline integrated into CI/CD (GitHub Actions + Docker),enabling real-time performance tracking and early drift detection across live forecasting outputs.
▸Containerised all ML services with Docker for reproducible training, testing, and inference environments across development and production, reducing environment-related failures to zero.
▸Applied PyTorch-based deep learning and ONNX model optimisation to energy time-series models, reducing inference latency for CPU-constrained production environments. Engineered a production-grade ETL pipeline processing Italian smart-meter energy data with multi-layer validation achieving 99% data integrity before model ingestion.

JAN 2021 – OCT 2023 · KOLKATA, INDIA

Data Analyst

@ LTIMindtree

▸Authored complex SQL pipelines for data extraction and transformation, reducing manual reporting preparation time by ~30% through automation.
▸Built and maintained Power BI dashboards for KPI monitoring used daily by cross-functional teams, replacing manual spreadsheet workflows with live data connections.
▸Conducted exploratory data analysis on large structured datasets, identifying data-quality issues and correcting upstream ETL logic.

[04]

Projects

proj_01

Multi-Tenant AI Customer Support Agent Platform (LLM / Agentic AI)

Production multi-tenant support agent built on LangGraph with per-tenant pgvector RAG isolation, verified for zero cross-tenant leakage under concurrent load (~53 req/s across 3 tenants). A confidence-based deterministic gate escalates low-confidence answers to human agents via webhook, with per-tenant usage metering and a live dashboard. Runs fully offline (local embeddings + extractive fallback) or against Postgres/pgvector; GitHub Actions CI and deployed live on Render.

LangGraph - RAG - pgvector - FastAPI - Docker - CI/CD - Render (live)

proj_02

Insurance Claims Triage & Risk-Scoring Agent (ML + LLM / EU AI Act)

Compliance-focused claims-triage system combining a deterministic policy verifier, a trained scikit-learn fraud classifier (fixed seed, top-factor explainability) and an LLM that drafts settlement reasoning while code computes the amount. A non-bypassable human-approval gate is the only path to settlement, backed by a fully queryable audit trail attributing every step to code / model / LLM / human. README maps the pipeline to EU AI Act high-risk controls.

Scikit-learn - LangGraph - Explainable ML - FastAPI - Docker - EU AI Act

proj_03

Document Intelligence & ERP Integration Agent (LLM / Automation)

Preprocessed 360K+ Amazon food reviews using NLP techniques (tokenization, stop-word removal, TF-IDF). Fine-tuned a transformer-based BERT/RoBERTa model to classify sentiment into positive, neutral, and negative, achieving an 88% F1-score and outperforming VADER and baseline models. Handled imbalanced classes, implemented multilingual detection, and served the model with MLflow plus live monitoring via Weights & Biases in a Docker-ready deployment.

LLM Extraction - Pydantic - OCR (Tesseract) - SAP OData -FastAPI - Docker - CI/CD

proj_04

Grid Anomaly Detection & Predictive Maintenance Pipeline (Energy / MLOps)

End-to-end MLOps pipeline for predictive fault detection on Italian electricity-grid sensor data (TERNA/GSE public datasets), pairing an LSTM sequence model with Isolation Forest to flag early anomaly signals. Implemented the full production loop - MLflow experiment tracking, Airflow automated retraining, Evidently AI drift monitoring, and a FastAPI inference service containerised with Docker and deployed via GitHub Actions CI/CD.

Python - LSTM - Isolation Forest - MLflow - Evidently AI - Apache Airflow - FastAPI - Docker - GitHub Actions

proj_05

GDPR-Guardian - PII Detection & Redaction Service (NLP / Governance)

Production-ready NLP service for automated PII detection and redaction in Italian/European text. Custom Microsoft Presidio recognisers detect Codice Fiscale and Italian-specific entities using spaCy's Italian model and custom NER. FastAPI gateway with Docker containerisation and an automated test suite in CI/CD - aligned with GDPR by Design and EU AI Act Article 10 datagovernance requirements.

Microsoft Presidio - spaCy (Italian NLP) - FastAPI - Docker - CI/CD - EU AI Act - GDPR

proj_06

Transformer Fine-tuning Pipeline - BERT / RoBERTa (NLP)

End-to-end pipeline fine-tuning BERT and RoBERTa on 360K+ text samples, achieving 88% F1-score on 3-class sentiment classification and outperforming VADER and TF-IDF baselines. Handled class imbalance, subword tokenisation, and multilingual input; full experiment tracking and model registry via MLflow and Weights & Biases. Deployed a Docker-ready inference endpoint; reproducible via fixed seeds, versioned datasets, pinned dependencies.

Hugging Face Transformers - BERT / RoBERTa - PyTorch - MLflow - Weights & Biases - Docker

proj_07

AI-Powered Commerce Agent

An AI-powered commerce agent built with LangGraph and the OpenAI API to automate product recommendations, order lookups, and policy-based cancellations. Implemented strict policy enforcement, traceable decision logging, and modular testing for intelligent and secure e-commerce interactions.

LangGraphOpenAI APIAgents

[05]

Education

2023 – 2026

Master's in Data Science

University of Napoli Federico II, ITALY

2018 – 2021

Bachelor of Computer Application

Asansol Engineering College, INDIAGrade 9/10

[06] — model.predict(next_role)

Let's turn data
into decisions

ejazfauzan14@gmail.com

+39 3421014345+91 9832887252MILAN, ItalyGitHub

[07]

Visitor Analytics

visits.count( )

total visits · 0 countries

live — geo via edge headers

visits.groupby("country")

collecting data — you are the first visitor ✦

FauzanEjaz