Agents in production

Assistants that can touch production

Tool calling, write safety, evals and cost control on .NET and Azure

The model decides which tool to call. What that tool may change, what the call costs, and whether it still behaves after a prompt edit all stay in deterministic server-side code.

LCP < 1.5s

Performance First

>99.95% SLA

Reliability

Multi-region

High Availability

What I work with

What I reach for, and what I have run in production.

Azure OpenAIAI Engineering
Amazon BedrockAI Engineering
Model Context ProtocolAI Engineering
LLM EvalsAI Engineering
Kubernetes (AKS/EKS)Core Infrastructure
TerraformCore Infrastructure
DockerCore Infrastructure
AWSCloud Providers
AzureCloud Providers
OpenTelemetryObservability
New RelicObservability
.NET 10Backend & Data
KafkaBackend & Data
PostgreSQLBackend & Data
Next.jsBonus
ReactBonus

Agents in Production

The gap between a working demo and a system you can leave running

A demo ends when the model calls a tool and something happens. In production the same agent works on real data, spends money on every call, and has to be stoppable before it does damage. The .NET ecosystem still has thin coverage for that.

Tool Calling

Tools defined over the domain services the product already uses, so the agent inherits their validation and permissions instead of reaching around them.

Write Safety

Reads stay open, writes stay bounded. Each tool declares what it can mutate; anything outside that boundary needs an explicit approval step before it runs.

Evals

A fixed set of cases replayed on every prompt, model or tool change, so a regression shows up in CI rather than in a customer's account.

Cost Control

Token accounting per run and per tenant, with the caps and fallbacks that keep an agent loop from turning into an unbounded bill.

The Work Underneath

The platform and delivery experience the AI work sits on

FinTech Platform Delivery

Built and evolved multi-tenant services processing high-frequency financial data, with delivery guarantees that hold through broker and consumer restarts.

.NETAzureKafkaRedisTimescaleDB

CI/CD Delivery Modernization

Moved legacy delivery pipelines to modern GitOps workflows with safer, repeatable deployments.

KubernetesHelmGitHub ActionsTerraformArgoCD

Got an assistant stuck before the write?

Assistants that can touch production | Alex Martínez