Tao Zhang

I have been working with a large scale software system that has a total of 14 million lines of code (C#/C++) for 11 years. It used to take 8-10 hours to build and generate the installation kit or deploy it to the cloud. With the old baseline build, it could only generate one build and one kit per day.

We built the CI (continuous integration) system in Python mainly and also used Powershell, Jenkins, and other technologies to implement automated processes in the development cycle.  After setting up the CI system, the average build time is about an hour and installation kit generation/deployment takes about 1-1.5 hours.

We also introduced many code metrics such as compile warnings, code complexity, and unit test coverage into the CI system. In addition, the CI system also helps to run all kinds of automation testing and generate reports. The CI system improved development efficiency for about 200 developers, QAs, and managers.

I redesigned many complicated classes in a large-scale C++ codebase following different design patterns, such as singleton, factory method, composite, adapter, or observer. The C++ codebase was refactored to make it testable by extracting to interfaces. I used GTest/Gmock provided by Google to test our legacy C++ class. 

As a large-scale real-time acquisition system, we set up a central place to store logs, performance data, dump files, etc., which are collected from different desktops or public cloud deployments.
We developed a web service to provide automated processes for different purposes, such as log file analysis to find the most popular tool string combination, performance data analysis to find out potential performance bottlenecks; dump file analysis to find the root cause of a crash or exception. We use C# with .NET Core and Python to develop the back end and Node.js and JavaScript to develop the front end.
SQL Server and Cosmos DB NoSQL as DB were implemented for data storage, and Redis cache and memory cache as another layer of data storage to achieve better performance. This web service was deployed to Azure.

I designed and implemented REST APIs and endpoints for a payment workflow, serving as an e-payment web service or a mobile app. The web service runs in Azure and communicates with different payment service providers to finish payments. It is built on top of C#, ASP.NET, and .NET Core. It utilized different layers of data storage: SQL Server, Cosmos, Redis Cache, and an in-memory cache in different stages of the whole workflow to achieve the best performance.

We use a Python script to perform the automation testing and build a CI/CD pipeline on top of Azure DevOps.

We built different web pages with Node.js and Angular. These web pages serve developer troubleshooting, statistics, and reporting for managers and operations staff, REST API testing, and client demos.

I developed a web service to provide different format files, such as DLis, or PDF, as the client requires deliveries of a large, real-time acquisition system. I used Java and Python as back-end coding languages to support a web service running in AWS. For the database, I used Cassandra and Amazon S3 as data storage. Further, I used Node.js for the front end to set up a simple UI and perform API testing. TDD is applied throughout the whole lifecycle of the project.

I built a 3G device management system running in Unix for Nortel Networks. That system was developed by Java together with a small part of a C++ component. The C++ code mainly communicates with 3G devices via socket connection based on a special protocol.
Java code builds whole systems based on MVC, including UI, business logic, and data models. I used JNI to connect Java MVC parts with C++ components to get and set the configuration of 3G devices from the UI down to the hardware.

I rewrote a Java-based REST API service using C#/.NET Core. This REST API server performs the following things:
• Receive XML requests from a financial organization
• Deserialize the raw XML request to a specific data model based on different request types in XML
• Apply different logic to convert data models in different workflows
• Serialize the final data model to XML and send it back

The client produces metrology software. The software allows for gathering precision measurements and allows its customers to test their measurement equipment to see if it is reliable. The client needs to create an Excel add-in from scratch that takes various inputs and then creates the Excel sheet template dynamically.

Created a call center platform to support different industries, providing the following common functionalities:

1. An IVR system integrated with Nortel PBX system to support the configurable automatic voice response system to provide information query, booking, etc.
2. An automatic fax system.
3. IP phone call.

I worked in the data security department to provide tokenization and key management service to support large volume transactions for internal customers in the bank. I built up a highly scalable, highly available, and high-performance web service that deploys to AWS and a private cloud (based on CloudFoundry) with Java and Spring Boot as the back end, Cassandra and DynamoDB as data storage, and Apache Kafka as a message queue. I also mentored new team members to catch up on project and technical stacks quickly.

The Authorization service is a highly available, high-performance, and scalable REST API solution built in Oracle OCI using Go for the back end.

The solution includes three core microservices:
1. Policy and Role Management: Manages roles, artifact types, data security conditions, and policies stored in Oracle DB.
2. Authz Decision Service: Delivers runtime policy decisions and maintains an in-memory store synchronized with Oracle DB.
3. Job Service: Schedules and monitors long-running tasks, and supports asynchronous APIs for importing large volumes of data into Oracle DB as the source of truth for the authorization decision service.

To improve development efficiency, the team implemented:
1. Using Codex with OpenAI GPT 5 to generate unit test code, integration test code using Go, Java, Python, Shell, etc.
2. Using Codex with OpenAI GPT 5 to automate the DevOps manual steps using Python, JavaScript, TypeScript, etc.
3. CI/CD pipeline scripts for deploying services to Kubernetes pods in Oracle OCI.

1. Cerbos Authorization Integration (Primary Focus)
• Integrated Cerbos Policy Decision Point (PDP) with the Python back end.
• Implemented authorization checks across the application.
• Created and maintained Cerbos policy files for resource-based access control.
• Developed database integration supporting both SQLite and PostgreSQL.
• Added feature flag protection for Cerbos functionality.

2. User Invitation System
• Implemented a complete user invitation system (back end and front end).
• Created user role assignment functionality.
• Fixed multiple issues with user invitation after merge/rebase operations.
• Updated database schema for user invitation tables.

3. Docker and Infrastructure
• Created Docker support for the Cerbos server.
• Fixed Docker Compose configuration for both demo and local environments.
• Updated Helm chart templates for Cerbos deployment.
• Created scripts for building and launching services via Docker.

4. Front-end Development
• Implemented user invitation UI components in Svelte.
• Created AcceptInvite.svelte component.
• Updated Users.svelte and InviteUser.svelte components.
• Integrated authorization UI elements.

I developed the script using Python and VBA to integrate Shopify and ShipStation with CorelDRAW to automate the entire process, from receiving the order to the wood shop owner making the product for the customer.

MAIN FUNCTIONALITIES:
• Extract the order information and notes from Shopify and Shipstation.
• Update the template in CorelDRAW to generate the design of the product based on the customer's needs.