Work History

Olostep is a web search, scraping and crawling API for AI. I own the cloud infrastructure behind its headless browser platform, which processes 1.2M+ requests a day across 7 browser engines.

• Own the production headless browser platform on GKE: each engine autoscaled by KEDA on queue depth, Terraform owning GCP, Helm owning Kubernetes, and Argo CD auto-syncing production with self-heal and prune. The fleet scales to 600+ spot nodes to absorb traffic spikes of 64x the median hour
• Own the Prometheus, Grafana, Loki and Alertmanager stack, and rebuilt alerting so it routes to Slack and actually gets looked at
• Cut compute cost 30% by pricing 13 machine families against the Cloud Billing Catalog API and shipping a spot machine ladder through GKE node auto-provisioning; built per-engine unit economics down to cost per 1,000 jobs
• Leading the migration of a production PostgreSQL database off Aurora Serverless v2 onto self-managed PostgreSQL 17 on local NVMe: measurement first, replay against a candidate box, durability before production data
• Led AWS-side deployments (ECS, ECR, ALB, DynamoDB, S3, CodePipeline), contributed to an open-source reverse proxy for client-based routing, and led API, SDK and MCP integrations with Zapier, n8n, LangChain, CrewAI and Airflow

AI sales agent for B2B distributors, processing quotes, deals and supplier communication across a multi-provider LLM pipeline on NestJS, Python, AWS and RabbitMQ.

• Rebuilt failure handling across NestJS and Python microservices so a provider outage or model change never took the pipeline down: circuit breakers, intelligent throttling and exponential backoff around flaky third-party APIs, an outbox pattern so a database write and the message announcing it could not diverge, and atomic job claiming that gave consumers exactly-once semantics
• Rebuilt the processing pipeline off a single-threaded bottleneck by making it idempotent and safe to run concurrently: row-level locking, thread reuse, and an event-driven design where each stage deploys to serverless on its own and scales horizontally
• Cut the manual review queue by 90% with a Pydantic validation layer and confidence scoring in front of the multi-provider LLM pipeline (OpenAI, Anthropic, Google) on LangChain and LangGraph. Processing accuracy went above 99%

Led engineering, system design, cloud and DevOps for a logistics platform serving Amazon and Compass Group, handling 2M+ API requests a month with real-time tracking, geo-fencing and live ETAs.

• Led a cross-functional team building a data-driven logistics and delivery platform with real-time tracking, geo-fencing, compliance checks and live ETAs. Route optimization work cut 30 to 60 percent of time, fuel and cost
• Led cost optimization initiatives, reducing geo-fencing costs by 70%
• Mentored a team of 10+ across engineering, business and UX, ran code reviews and build-versus-buy calls, and owned the cloud infrastructure. Time from merge to production went from 4 weeks to 1 week
• Defined technical strategy in direct partnership with leadership from Amazon and Compass Group

Full engineering on gMeets, a large-scale meetings platform: .NET Core services and the backend behind them, plus cloud infrastructure, CI/CD, infrastructure as code, automated and load testing, security, and database performance.

• Built and maintained enterprise .NET Core services for the meetings platform, from the backend logic through to the messaging on RabbitMQ
• Owned the CI/CD, Terraform infrastructure as code and security automation around those services, improving deployment efficiency by 40% on Azure
• Increased SQL data retrieval speed 20 to 40% through query plan analysis and asynchronous patterns in the application
• Added Selenium end-to-end tests that removed 40% of manual testing effort, and ran JMeter load testing with Groovy-scripted scenarios that improved performance and scalability by 20%
• Integrated SonarQube into the build, addressing 20+ security vulnerabilities along with cognitive complexity and dependency issues

End-to-end engineering across .NET, Azure, AWS, React, TypeScript and Python, plus DevOps, Linux automation, infrastructure as code, and cloud architecture for international clients.

• Ran a self-hosted on-prem Docker-enabled Linux server farm of 20+ machines at 128 GB RAM each, roughly 2.5 TB of combined memory, with automated Android emulators and custom provisioning scripts
• End-to-end cloud automation on AWS and Azure: CodePipeline, ECS, ECR, ALB, DynamoDB, S3, plus Azure private services, vNETs and VPC peering, with GitLab and GitHub automation on self-hosted containerized Linux instances
• Expanded an in-house point of sale platform onto .NET microservices with Docker, Terraform, GraphQL, Grafana, Prometheus and CloudWatch, improving performance and reliability by 30%
• Refactored a legacy application to 99% test coverage with unit and integration tests

Delivered core modules of vava.cars, a large-scale microservices product for a Fortune 500 client, on .NET Core, Angular, SignalR and Azure.

• Built four core modules on a $50M Series B microservices product, including a real-time notification and bidding engine on SignalR
• Worked with the incident team on the failures that came with scaling across a server farm: HTTP connections held open instead of released, database connection exhaustion fixed with a caching layer, and session affinity so a real-time session survived landing on a different node
• Optimized database queries and implemented async processing for bulk operations, with up to 13x gains on SQL Server queries

Early career building .NET applications for the largest credit union service organization in the US and a drag-and-drop website builder.

• Re-architected a credit union web application using ASP.NET MVC, Web API and Entity Framework with DDD principles, improving search efficiency by 60% through Elasticsearch
• Built a drag-and-drop website builder for lead collection with a custom DOM undo/redo library