Bagrat Aznauryan

Developed an HTTP API using Python and Flask, which several machine-learning models backed. I utilized Python and an SDK library and worked on back-end APIs, DevOps processes, using Kubernetes and AWS, and distributed scraping.

After a successful acquisition of another SaaS company and its technology, Sauce Labs needed to have the migrating customers have a transparent and smooth transition to a new system and allow old customers to start using the new technologies.

Operating in the testing automation industry, the main entry point for a customer into the testing cloud system was WebDriver-compliant services for both existing and acquired companies.

The project I worked on was implementing a new facade web service that should route the WebDriver requests to the matching system (old or acquired). It had to be distributed, scalable, and fault-tolerant since going down meant no service.

I have designed and implemented this service with multiple caching layers (local memory, a warm-up based on a message queue, and Redis for a fallback). The service code was 90% covered by tests, allowing a confident and quality development process.

The service beta went into production in three weeks with a gradual migration of the customer, which helped keep existing customers of the acquired company and increased revenue by having old customers start using new features and capabilities.

I implemented a distributed data scraper integrated with a VPN client with auto-rotating and expiring VPN configurations per IP. This was implemented using a combination of Python asyncio, Redis, and Docker Compose.

The scraper collected more than 20 million company data into a structured format saving the company hundreds of thousands of dollars required to buy such a commercial database.

I worked on migrating multiple services from a manual VM deployment using Fabric (Python) to a fully automated CI pipeline-triggered, Helm-based deployment to Kubernetes at Sauce Labs Inc.

The project involved defining Kubernetes resources with appropriate deployment strategies, packaging into Helm with proper versioning, and releasing Helm Charts in an automated GitLab CI pipeline. As a follow-up project, all the services code was instrumented for monitoring using Prometheus clients, and Grafana dashboards were built to make monitoring easy and insightful.

This increased the release/deployment frequency, reduced deployment failures, made it a lot easier to debug failures, and significantly reduced the stress of making a deployment.