top of page

Portfolio

Marketing Mix Modelling

Marketing Mix Modelling

- Created MMM bayesian models for a pharma company
- Worked on infrastructure related topics to deploy at scale models

Env & tools: AWS, STAN, Meridan, Kubernetes

Precision marketing - Targeted advertising (Scoring/big data)

Precision marketing - Targeted advertising (Scoring/big data)

Developed multiple appetency scoring algorithms on Spark Databricks for Samsung to predict website visitors interests and target them with the right marketing campaigns based on online navigation and CRM data

Env & tools: Azure, databricks, spark

Audience segmentation (analytics/SQL)

Audience segmentation (analytics/SQL)

We analyzed our visitors data base and provided a methodology to split them into segments showing the value of each to the business.

ENV & tools: Big query, python

Customer analytics and segmentation in the travel industry (analytics)

Customer analytics and segmentation in the travel industry (analytics)

Worked on quite varied use cases shaping the future of travel and tourism in Dubai (Insights about tourists’ behaviors, time series, forecasting, data clustering and markets segmentation..).

Env & tools : dataiku, python

Customer behavior analysis and churn prediction (analytics)

Customer behavior analysis and churn prediction (analytics)

Used varios data sources on big query (pricing, reviews, transactions, ...) to get the CLV, the reasons behind customer insatisfaction and churn for a big e-commerce actor.

Env & tools: GCP, big query, python, seaborn (viz)

Realtime information retrieval in call centers (NLP)

Realtime information retrieval in call centers (NLP)

The goal of the project was to help call center agents handle customers requests efficiently by providing them with the right tool to query a knowledge base and find solutions. For this, I developed a NER model that extract keywords from the call transcript in real time and retrieves the best-matching articles.

Env and tools: Spacy, prodigy, python

ML models deployment at scale (ML ops)

ML models deployment at scale (ML ops)

Deployed ML models at scale on Kubernetes using MLflow and performed load tests to address scalability and availability issues.
Check out my medium articles describing this project - https://tinyurl.com/3nfr8w24

Env & tools: GCP, k8s, docker, mlflow

Real time market sentiment analysis dashboard (ML ops/dashbording)

Real time market sentiment analysis dashboard (ML ops/dashbording)

Using twitter API and a custom home made yahoo finance scraper to get stock tweets and prices. For real-time processing we opted for Kafka and used Bert as a sentiment analysis model. The front was built with streamlit and connected to postgres database.

Env & tools: kafka, postgres, python, streamlit

Demand Forecasting (time series)

Demand Forecasting (time series)

Implemented time series models to forecast demand in the retail sector combining various inputs (promotions, sell-in, sell-out...)

Text mining and insight generation (NLP)

Text mining and insight generation (NLP)

Used text mining techniques to analyze millions of reviews collected
by the french government to understand citizens demands.

Env & tools: python, spacy, hugging face

Data tables design and dashboards creation (BI)

Data tables design and dashboards creation (BI)

A two-month project in which I designed the tables and created dashboards on google data studio to monitor marketing campaigns performance.

Env & tools: datastudio, python

Customer choices analysis (analytics)

Customer choices analysis (analytics)

Worked on choice modelling within airlines and travel industry.
Performed choice analysis and customer behaviour modelling through observed trade-os when buying airlines tickets.

Env & tools : flask, python

Text localization and recognition in images using deep learning (DL)

Text localization and recognition in images using deep learning (DL)

Implemented a two-step pipeline detect and recognize texts in image.
We finetuned and tested state of the art deep learning models such as single shot multibox detector, yolo... and a seq-2-seq model with attention layers for the recognition task.

Env & tools: on premise, docker, tensorflow (keras)

bottom of page