What we've built

Challenge

The client wanted to automate phone support on one non-negotiable condition: no conversation could leave their network.

Solution

We built the agent on top of their phone system: speech recognition, an open language model and speech synthesis on their own GPU, with audio processing, natural interruption, search across their documentation and traceability for every turn.

Result

Real calls handled end to end with responses in under 2 seconds, in natural Spanish and with all data kept inside the client's network.

Challenge

A data platform maintained by more than 20 engineers was handed to a small team after a company restructuring.

Solution

We took over the platform and led the team: real-time data ingestion on AWS, validation and routing of time series, detection of critical events and alerts, a data lake for the data science team and an AI assistant on the live data.

Result

The platform kept running and evolving with a fraction of the original team.

Challenge

Moving a platform with more than 100,000 users to another cloud provider without disrupting any of its clients.

Solution

A new platform on Kubernetes in Azure, with all infrastructure defined in Terraform and automated deployments. Migration one client at a time.

Result

In progress, in phases: each client is migrated and validated independently.

Challenge

Every deployment of the platform was manual and took a week's work.

Solution

We moved all the infrastructure to Terraform and automated the deployments in Azure DevOps.

Result

Deployment time dropped from a week to an hour.

Challenge

A product team wanted to add AI to its software with no previous experience of taking it into production.

Solution

We chose and introduced the platform (AWS Bedrock with Claude models), with traceability and evaluation in Langfuse, and set up the working patterns: tool calling, structured output, prompt versioning, content and personal data filters, and regression tests.

Result

AI in production in two months, on a foundation the team adopted.

Challenge

An agency with many internal applications, no in-house technical team and undetected security risks.

Solution

An audit of all their software, fixes for the most serious security flaws, consolidated infrastructure to cut costs, and a spec-driven development workflow with AI agents, reviews and checks.

Result

Critical risks fixed, infrastructure unified and a non-technical team able to evolve its own software with AI agents.