Ana G.

Senior .NET Engineer | AWS & Azure | Elasticsearch | AI: RAG, Evals

I build .NET systems on AWS and Azure that hold up in production, including Azure OpenAI features, and I bring the same rigor to AI work: RAG measured against a test set, and AI-assisted delivery with Claude Code. 10+ years, 7 of them on AWS.

Good fit if you need:

• AI features inside an existing .NET product on AWS or Azure (LLM endpoints, RAG over your data, assistants)
• An AI feature made measurable: test sets, retrieval metrics, an AI judge checked against human ratings
• Data-heavy .NET backends: APIs, ETL, search, background processing
• A secure cloud setup for AI: IAM on AWS, Entra ID and managed identities on Azure, secrets management, cost control

RECENT WORK: MULTI-TENANT E-COMMERCE SAAS ON AZURE (.NET 10)

• Azure OpenAI (GPT-4.1) in production, generating customer-facing content
• Owned a batch processing pipeline on Azure Functions and the webhook-driven, queue-based integration with a shared platform
• Idempotent, recoverable workflows over multi-tenant Cosmos DB, zero-downtime schema evolution, resilient batch writes to a third-party API
• Security design for a shared platform: Entra ID between services, Entra External ID for partners, managed identities with access keys disabled, deployments without stored secrets
• Cut projected non-production Azure spend by 57-69% (right-sizing, KEDA scale-to-zero, a liveness-probe redesign that ended restart loops)
• Container Apps, Service Bus, Redis, Key Vault, Terraform and Bicep

AI ENGINEERING

• AI-assisted delivery: led the team's move to Claude Code with spec-driven development (Spec Kit), an independent reviewer agent and review-fix-re-review loops in isolated git worktrees. Built the guardrails: custom skills and commands, hooks that block unsafe actions such as pushes to main, and CLAUDE.md rules that codify the architecture.
• RAG with evaluation, personal project (.NET, OpenAI, Qdrant): parent-document retrieval, metadata boosting by votes and recency, and a 100-question test set scored for retrieval (Hit@K, MRR) and by an AI judge. I keep only what measurably helps;
• Agent evaluation: trained and evaluated code-focused AI agents on real Python repositories (Black, Dash) for Stellar AI.
• Local RAG on AWS Bedrock with Titan embeddings and OpenSearch, personal project.