Ahmed – AWS, Python, Terraform

Ahmed contributed to PyTorch Geometric, a leading open-source graph machine-learning framework, by building infrastructure that connects relational databases with heterogeneous graph-learning models.

• Contributed to the PyTorch Geometric (PyG) open-source framework by addressing and resolving GitHub issue #9839;
• Implemented targeted bug fixes and performance enhancements for graph neural network data structures;
• Wrote comprehensive unit tests to ensure code stability, regression prevention, and cross-version compatibility;
• Collaborated with core maintainers and open-source contributors through code reviews and technical discussions;
• Optimized data processing workflows and pipelines for graph-based machine learning models.

Ahmed worked as an independent Data and AI Consultant through Innosol Online OU and was engaged via UpWider to take ownership of a stalled portfolio of enterprise AI and data-analytics initiatives at the National Bank of Belgium. The engagement operated within a highly regulated financial environment with strict security, privacy, and operational-resilience requirements.

• Architected AI-driven analytics solutions;
• Took operational ownership of a complex enterprise backlog covering AI, financial reporting, risk analytics, and data-platform initiatives;
• Architected AI-powered risk-analytics systems and optimized high-throughput Delta Lake pipelines on Databricks;
• Integrated private LLMs into controlled Azure and Databricks environments to support secure financial reporting;
• Automated financial reporting and risk-analysis workflows while complying with GDPR and DORA requirements;
• Cleared critical project backlogs, accelerating the institution’s analytics and digital-transformation roadmap.
• Optimized Delta Lake workflows on Databricks to support large-scale regulatory and risk-analytics use cases;
• Designed scalable data-processing workflows for complex analytics requirements;
• Improved data workflows to support reliable regulatory and risk-analysis processes.

A secure, AI-powered decision engine that combines LangChain and LangGraph agents with private LLMs to automate complex decision-making workflows. The product supports controlled, end-to-end processing of decision logic while keeping sensitive data and AI processing within a private environment. Ahmed worked as an independent Data and AI Consultant leading an end-to-end AI transformation for a financial payment-systems company. He translated operational bottlenecks in client onboarding, customer communication, and decision-making into secure automated workflows.

• Led requirements discovery and scoped the company’s end-to-end AI transformation roadmap;
• Architected and deployed automated client-onboarding pipelines and inbound customer-communication workflows;
• Combined LangChain/LangGraph agents with private LLMs for secure, automated decision-making;
• Designed agent workflows to support complex decision-making processes;
• Integrated private LLM capabilities into automated decision workflows;
• Reduced manual operational work across client management and customer communication;
• Improved decision accuracy and transaction-processing speed through controlled agent orchestration;
• Structured the decision engine for secure and reliable end-to-end processing.

Ahmed served as a Lead Machine Learning and Data Engineer at Blue.cloud, acting as the primary Snowflake subject-matter expert and client-facing consultant for Zendesk. He led enterprise AI and data-platform initiatives across complex multi-cloud environments.

• Led the migration of petabyte-scale data pipelines from GCP to Snowflake across a multi-cloud enterprise environment;
• Developed GPT-powered assistants in Snowflake Cortex, enabling non-technical stakeholders to query enterprise data using natural language;
• Directed cross-functional engineering squads through architecture, testing, and deployment of low-latency LLM orchestration workflows;
• Reduced query-execution overhead and infrastructure resource consumption through Snowflake architecture optimization;
• Reduced manual reporting effort by providing business users with direct natural-language access to governed enterprise data.

A fraud detection and real-time data processing solution designed to identify suspicious activity and support high-volume analytics. The product combines Azure ML with Delta Lake and Apache Spark to process streaming data at scale and enable timely, data-driven fraud detection.

• Developed robust fraud detection systems to identify and mitigate suspicious activity;
• Developed high-throughput streaming data pipelines utilizing Azure ML, Delta Lake, and Apache Spark to process large-scale data;
• Optimized data processing workflows to support scalable fraud detection and real-time analytics;
• Integrated machine learning capabilities into data pipelines to support automated fraud analysis;
• Improved data processing efficiency and reliability for high-volume streaming workloads.

An NLP-based entity resolution solution designed to improve master data management by identifying and matching records that refer to the same real-world entities. The product uses BERT embeddings and Random Forest models within production ETL pipelines to support scalable and reliable data matching.

• Designed production ETL pipelines for NLP-based entity resolution;
• Utilized BERT embeddings and Random Forest models to enhance master data management;
• Developed scalable data processing workflows for entity matching and resolution;
• Integrated NLP-based models into production ETL pipelines for automated data processing;
• Optimized data workflows to support reliable and consistent master data management.

A generative AI-powered ad-targeting solution designed to improve audience segmentation and enable more relevant advertising. The product combines GPT-2 text generation with Neo4j-powered Graph Neural Networks (GNNs) to analyze audience relationships and support deeper, data-driven targeting.

• Architected generative AI ad-targeting systems;
• Engineered data pipelines integrating GPT-2 text generation with Neo4j-powered Graph Neural Networks for deep audience segmentation;
• Developed scalable data workflows to support AI-driven audience analysis;
• Integrated graph-based analytics with generative AI capabilities for targeted advertising;
• Built and productionized a GPT-2 personalization system using a Twitter relationship graph of approximately 3 million users to identify micro-influencers and generate targeted advertising content;
• Deployed nightly batch inference on AWS V100 GPU instances with autoscaling and load balancing;
• Implemented Bitbucket CI/CD, S3 model versioning, production metrics, and custom monitoring and evaluation scripts;
• Optimized data processing workflows for complex audience segmentation use cases.