Muhib Ullah Khan

A Tecton-based feature store to serve high-quality, reliable, and reusable features to the downstream ML applications. I was responsible for migrating the existing data pipelines and building new features into the feature store from various batch and stream sources. Once the product matured as a centralized feature repository and was widely adopted across the company, I built strict feature quality, observability, and governance standards around the use of the product. In this project, Tecton served as the base platform while Databricks and PySpark were the main components for building feature engineering pipelines.

A Soda-based data quality framework to simplify and automate the creation of data quality checks on critical data assets. I was involved in building this framework from the ground up and ensured its timely delivery to our internal customers. This framework automated data freshness and completeness checks and simplified the creation of custom checks by providing a Python interface. This enabled engineers and data scientists to easily write DQ checks on their data assets. This framework was built in Python and used Soda as the back-end DQ engine. The solution also included orchestration, scheduling, alerting, and observability of DQ checks. The technologies used were Soda, Python, Prefect (for orchestration), Soda UI (for observability), AWS SNS, SQS (to send raw check results to storage), and integration with Slack, Opsgenie (for alerting).

As part of the platform team, I was one of the owners of multiple Databricks workspaces and was involved in governance, maintenance, and admin-related activities. Databricks was integral to my team's work and served as the backbone for our operations across the tech domain. I regularly ran internal support to resolve user queries on policies, permissions, error resolution, data, and storage access, as well as provided solutions suitable to their needs.

This data analytics project reports on the performance of a famous car manufacturing brand and how it competes with other brands in different European countries. In this project, I implemented the data ingestion process from landing to a raw zone, the raw zone to Operational Data Store (ODS), and finally from ODS to the data warehouse. I used Snowflake as the target data warehouse, DBT to write transformation scripts, and Azure services such as ADLS, PySpark, and Databricks to implement data ingestion and data lake storage.

This real-time monitoring app refreshes every minute to visually represent the current status of machines working in a factory. I oversaw ingesting real-time data from IoT devices using Azure Event Hub, Spark, or Databricks, loading the data in a SQL database, and performing further transformations and aggregations to prepare the data for visual representation on the dashboard.

As a data conversion specialist, I worked on this SQL-based project. I communicated with business analysts to understand business requirements and then map source data to appropriate fields in the target database. Also, I documented data mappings and transformations required for conversion and wrote conversion scripts to extract, transform, and load data into the target database.

This machine-to-machine (M2M) service for eSIM users, usually companies, monitors and controls their devices connected to a telecommunication network. In this project, I oversaw automating the billing process so that when the real-time usage data would come in from all the connected devices, it could be processed through a series of scheduled SQL jobs and summarized in the form of an invoice. While working on this project, I was also able to identify and fix a bug in the code that was causing the company to lose hundreds of dollars per invoice for large customers.

This data analytics project for a restaurant visualizes the performance of its food delivery partners and analyzes factors affecting delivery on time. In this project, I oversaw designing a data mart and then loading the restaurant's data in it. I also created a Power BI dashboard for the client to quickly visualize how well the deliveries are going.

An agentic, multi-modal application leveraging large language models (LLMs) to automate high-quality text and image generation, eliminating manual photography workflows and accelerating product time-to-market. The system was architected as a highly concurrent, scalable, and stateless API that seamlessly integrated with the core application to eliminate manual content-creation bottlenecks.

I was responsible for designing and scaling the application's underlying data infrastructure. This involved curating historical enterprise data to build a high-fidelity context layer, implementing a multi-modal Retrieval-Augmented Generation (RAG) system, and optimizing prompts to ensure accurate model outputs. I also engineered a specialized, high-throughput asset store for approved and AI-generated imagery, creating a unified source of truth that served both multi-modal model inputs and downstream evaluation pipelines. Finally, I leveraged application logs and model responses to build an observability pipeline that generated content quality metrics, establishing a continuous feedback loop that maximized generation efficiency and minimized API costs.