Data Engineer Intern & Data Science Intern - PFE 2027

About us:

At Cognira, we strongly believe that people are the biggest asset of our company. Our hand-picked team consists of passionate, collaborative, and forward-thinking individuals from all over the globe. We are passionate about making science easy and accessible to retailers, helping them get more value from people, data, and systems. We bring together expertise in retail, science, and scalable technologies to automate and enhance the quality of decision-making through software and consulting services.

We are proud to have a growing team of domain experts and data scientists, as well as a culture that fosters strong and long-lasting relationships with our clients.

Our Locations:

Atlanta, GA, US

Tunis, Tunisia

Istanbul, Turkey

Paris, France

Porto, Portugal

Our values:

Be nice and trustworthy

Put our customers first

Succeed together

Never stop improving

Are you ready to grow with us? To find out more about Cognira, please visit our website at www.cognira.com

Please submit your resume in English.

I- Data Engineering Internship Topic:

Topic 1: Data Pipelines with DuckDB

Cognira | Data Engineering Team

ABOUT THE INTERNSHIP

Our data pipelines run every transformation with Apache Spark on cloud clusters, even when the data is small enough for one machine. This project moves suitable transformations to DuckDB, a fast SQL engine that runs on one machine, and lets them run on a developer’s laptop as well as in the cloud. The goal is faster and cheaper runs, and a shorter cycle to develop and test.

WHAT YOU'LL WORK ON

A short comparison of Spark and DuckDB on a few existing transformations, used to select which ones to move.

DuckDB versions of selected transformations, with automated tests that prove they give the same output as the Spark versions.

A tested way to run DuckDB transformations on Databricks, our cloud data platform, including the compute type, the dependencies and the access to our data. If Databricks does not support this directly, the project delivers a workaround.

Changes to our pipeline orchestration, the system that defines and schedules each step, so that each step declares whether it runs on Spark or on DuckDB.

One command that runs a DuckDB transformation on a laptop, with sample data or with data read directly from cloud storage.

One version of the code for each transformation, which runs both locally and on Databricks without changes.

WHAT YOU'LL BRING

Python, SQL

NICE TO HAVE

Databricks, Pyspark, DuckDB

II- Data Science Internship Topics:

Topic 1 : Semantic Similarity

Cognira | Data Science Team / Cognira

ABOUT THE INTERNSHIP

Use an LLM to read product descriptions and assess with they should be considered substitues, complements, "super-SKU" components, or how they should fit into a product hierarchy together.

WHAT YOU'LL WORK ON

A science module that groups products into likely complements, substitutes and assembles a product hierarchy based on reading the descriptions of products.

WHAT YOU'LL BRING

Data Analytics, Statistics, Python

NICE TO HAVE

Databricks, Pyspark, Knowledge of LLM techniques, Natural Language Parsing familiarity

Topic 2: AI Agent for Forecast Debugging

Cognira | Data Science Team

ABOUT THE INTERNSHIP

Use an LLM agent to investigate differences between our batch and real-time forecasts: it explores the data, configurations and features with read-only tools, and proposes the most likely root cause with supporting evidence.

WHAT YOU'LL WORK ON

A prototype debugging agent, its toolset, and an evaluation framework measuring how often it finds the correct root cause on known issues.

WHAT YOU'LL BRING

Python, Data Analytics, Software Engineering

NICE TO HAVE

Databricks, PySpark, Scala, LLM agents

Topic 3: Deep Learning for Promotional Demand Forecasting

Cognira | Data Science Team

ABOUT THE INTERNSHIP

Evaluate modern deep learning forecasting approaches, including pretrained time-series foundation models and transformer-based architectures, against existing production forecasting models. Investigate how incorporating richer product-level information can improve forecast accuracy, particularly for products with limited promotional history.

WHAT YOU'LL WORK ON

A reproducible benchmarking framework comparing deep learning models with production models, along with a detailed analysis of scenarios where deep learning methods improve promotional demand forecasting performance.

WHAT YOU'LL BRING

Machine Learning

Deep Learning

Python

Statistics

Time Series Forecasting

Databricks

PySpark

PyTorch

Large Language Models (LLMs)