Vadim Ovcearenco

I designed and implemented the AWS infrastructure for a skin image analysis platform that processes photos submitted by participants testing skincare products. The workload had highly concentrated peaks, with approximately 10,000 images arriving in the morning and another 10,000 in the evening, while individual image analysis took approximately one minute.

I also built a containerized processing pipeline using Amazon ECS and S3 around a machine-learning model provided by the data science team. The infrastructure was designed to scale processing capacity during the morning and evening peaks and scale back down as demand subsided, balancing processing throughput with infrastructure cost. I migrated the application from an earlier GCP implementation to AWS and adapted the processing architecture to AWS services.

I designed and built ArchPilotAI, a full-stack demonstration of safe, action-oriented AI architecture. It converts natural-language infrastructure requests into typed tickets and executable tasks, while deterministic policies—not the LLM—authorize or deny each action. The workflow adds human approval for sensitive changes and executes only validated operations through constrained MCP tools. I also implemented externalized YAML architecture knowledge, a FastAPI back end, a React interface, and unit, integration, end-to-end, live, and adversarial tests covering safety boundaries and prompt-based bypass attempts.

• Served as the sole back-end developer for an AI-powered aquarium design platform, taking ownership of substantial back-end refactoring and designing the integration with RunPod and ComfyUI for Stable Diffusion image generation. Built a workflow execution pipeline based on JSON graph definitions and integrated it with the Django back end.
• Improved the GPU development infrastructure by moving approximately 5GB of AI models from the Docker deployment path to network storage, reducing environment startup time from 5–7 minutes to approximately 30 seconds. Previously evaluated and configured an AWS EC2-based GPU deployment, including NVIDIA driver setup, before the project moved to RunPod.
• Implemented and maintained Django REST APIs, Auth0 authentication, PostgreSQL integration, and AWS-hosted back-end components while collaborating with the front-end team.

I joined the project to investigate an intermittent issue where the system could freeze while calculating quiz winners. Analysis showed that the existing vote-processing implementation would not operate reliably under the required burst load, so I redesigned and rewrote the flow for storing and processing results.

I implemented the critical path using low-level Redis operations to efficiently register and process approximately 10,000–20,000 participant responses arriving within a 2–3 second window while preserving the response timing required to identify the fastest correct winner. The redesigned implementation was subsequently used successfully at a live stadium event, as confirmed by the client.

I developed the email ingestion and archiving subsystem for a wealth management platform used to preserve client communications and web data.

I designed and implemented the workflow for receiving incoming emails through AWS services, processing SNS notifications, retrieving email data from S3, parsing messages and attachments, and persisting the resulting data for archival and downstream processing. I focused on making the ingestion workflow robust and maintainable, including handling different email structures and attachments and integrating the processing pipeline with the existing Django application.

I took over an existing commercial real estate financial modeling platform as the primary back-end engineer and brought it to production. I worked directly with the client, independently maintaining and extending the Django application, resolving issues, and adding unit test coverage to improve reliability.

I extended the Excel-based financial modeling functionality using PyWin32 and COM automation, including evaluating alternative approaches to Excel integration against the project's requirements. I built deployment automation for the AWS-hosted application and continued to support and extend the system after taking ownership from the previous developer.

Worked as the key back-end developer on an interactive live quiz platform designed for thousands to tens of thousands of concurrent participants. Viewers watched questions broadcast on television and submitted their answers through the web application, requiring a low-latency real-time communication layer.

Designed and developed the Django back end and the real-time WebSocket architecture using Elixir and Phoenix. The Django application handled core application data and management functionality, while Phoenix WebSocket channels supported real-time communication with large numbers of connected clients.

The system was designed and tested for high concurrency, although the product did not ultimately reach production.