Hello, I'm

Farjad Ahmed

I build |

Data Scientist at HelloFresh SE, Berlin — delivering Bayesian MMM models for €1B+ marketing spend across 18 markets, with a research background in transformer optimization.

Get in Touch
0+ Years Experience
0 Markets Optimized
0B€+ Spend Modeled

Scroll Down

01. About Me

I'm a Data Scientist with a research background in transformer optimization and hands-on experience building Bayesian Marketing Mix Models at scale. Currently at HelloFresh SE in Berlin, I deliver hierarchical MMM models using PyMC for €1B+ marketing spend across 18 global markets.

My Master's thesis at the University of Hildesheim focused on building robust and compact transformers for dense prediction tasks, achieving a 17% parameter reduction in Conditional DETR while maintaining competitive performance on COCO.

I thrive at the intersection of rigorous statistical modeling and production engineering, from prior/posterior analysis and lift test calibration to Airflow pipelines and Databricks deployments.

Berlin, Germany
English C1 · Deutsch A2
MSc Data Analytics (GPA 1.6 / 5.0)

What I Do

  • Bayesian Marketing Mix Modeling
  • Computer Vision & Transformers
  • Hierarchical Statistical Models
  • Production ML Pipelines
  • Data-Driven Attribution

02. Projects

Featured Work

More Projects

View All on GitHub

03. Experience

Data Scientist

HelloFresh SE

Jan 2025 — Present Berlin, Germany
  • Developed and maintained hierarchical MMM models using PyMC for global optimization across 18 markets (€1B+ annual spend), tuning parameters and saturation curves through systematic prior/posterior fitting analysis
  • Monthly FIRM forecast delivery for MMM model across all entities and markets, coordinating staging validation, lift test calibration, and production deployment
  • Designed data-driven attribution (DDA) baseline using custom weighted attribution methodology (campaign-level cost, impressions, clicks, conversion ratios)
  • Integrated Lift Test calibration into MMM model, increasing stakeholder confidence and aligning predictions closer to ground truth
  • Onboarded new Data Science team members and trained them on the MMM model and data-driven attribution methodology
  • Evaluated alternative MMM architectures (hierarchical vs flat) using systematic hyperparameter optimization with Optuna to improve lift test calibration and inform architectural decisions
  • Proposed architectural review, improvements transitioning from sequential to joint parameter estimation, reducing model instability and addressing lift test propagation issues; documented model assumptions, edge cases, and architectural trade-offs
PyMC Bayesian Airflow Optuna Databricks

Data Science Intern

HelloFresh SE

Aug 2023 — Dec 2024 Berlin, Germany
  • Redesigned Streamlit dashboard for Media-Mix Conversions Model serving 16 markets, improving ease of use and user-driven insight visibility
  • Conducted ad-hoc analyses on customer acquisition costs, activation predictions, and adstock priors
  • Fine-tuned model priors using cross-market analysis and optimization, achieving <10% error between predicted and test values
  • Onboarded new markets through exploratory data analysis and model prior optimization
  • Maintained the model in production alongside a senior data scientist
Streamlit Python MMM

Data Product Manager (Werkstudent)

HelloFresh SE

Aug 2022 — Aug 2023 Berlin, Germany
  • Led Dimensions Modernization initiative for foundational fact and dimension tables used company-wide across all business units
  • Conducted customer discovery interviews, PRD documentation, and legacy asset deprecation
  • Conducted sprint reviews and maintained data asset documentation in Global Business Intelligence (GBI)
  • Coordinated cross-functional teams for data product development and deployment
Product Data Agile

Associate Product Manager

Astera Software

Jul 2021 — Feb 2022 Karachi, Pakistan
  • Led team of 3 developers and 1 QA engineer for data preparation tool product development
  • Led sprint plannings, reviews and project demonstrations alongside product research
  • Devised a proprietary scripting language for data cleansing, integrated with the product for data transformation and cleansing
  • Designed data cleansing components, UX and UI for the product
Leadership Product Dev UX/UI

Data Integration & QA Engineer

Astera Software

Sep 2020 — Jul 2021 Karachi, Pakistan
  • Developed automated and manual test cases for quality assurance, covering functional, regression, and system testing
  • Conducted data validation of ETL pipelines and reporting outputs, identifying and resolving issues with developers
  • Created R, Stata, and Python scripts for machine learning and statistical analysis components
R Python Stata

04. Technical Arsenal

Bayesian & Statistics

PyMC Hierarchical Models MMM Prior/Posterior Analysis Lift Test Calibration ArviZ

Computer Vision & DL

PyTorch TensorFlow DETR / Conditional DETR Vision Transformers Multi-Head Attention OpenCV

Programming

Python NumPy Pandas Scikit-learn PyCaret C/C++

Data Engineering & MLOps

PySpark Databricks Airflow DAG Management Multi-Stage ETL Pipelines AWS S3 Staging/Live Deployment

Research & Optimization

Model Optimization Parameter Reduction Ablation Studies Gradient Analysis Hyperparameter Optimization COCO API

Data & Analytics

SQL EDA A/B Testing Time Series NLP Big Data

05. Research

Master's Thesis

Building Robust and Compact Transformers for Dense Prediction Tasks

University of Hildesheim · 2023 — 2024

Implemented Dynamically Composable Multi-Head Attention (DCMHA) in Conditional DETR for object detection. Designed a custom 5-branch attention head composition architecture with shared feedforward networks, achieving 17% parameter reduction while maintaining competitive performance (AP50:95: 39.3 vs 39.2 baseline on COCO).

17% Parameter Reduction
39.3 AP50:95 on COCO
5 Branch Architecture
PyTorch Conditional DETR Vision Transformers DCMHA COCO Pascal VOC
Read Thesis

06. Education

Master of Science in Data Analytics

University of Hildesheim

Apr 2022 — Dec 2024 · Hildesheim, Germany

GPA: 1.6 / 5.0 (German) ≈ 3.7 / 4.0 (US)

Advanced Computer Vision Advanced ML NLP Optimization Big Data Analytics

Bachelor of Science in Electrical Engineering

Habib University

Aug 2015 — Jun 2019 · Karachi, Pakistan

Thesis: "Multi-Agent Robotics Teaching and Research Platform"

07. Get in Touch

I'm always open to discussing data science, Bayesian modeling, research opportunities, or just connecting. Feel free to reach out!

linkedin.com/in/ahmedfarjad github.com/FarjadAhmed
Connect on LinkedIn