Skip to content

Wendelmaques · Infrastructure consulting

AI in production.

Infrastructure from planning to operations.

I design and operate AI platforms for your company. I connect data, models and systems, with access control, cost control and recovery.

From data to product

Documents, a server rack and a desk with a screen and telephone connected by a red route.
  1. Data
  2. Processing
  3. Application

The model is one stage. Infrastructure connects company data to the product and supports operations.

AI-generated editorial illustration.
12+ years in engineering
Software, telecom and system operations
AI, data and speech
Platforms, sensitive data and existing systems
Work directly with me
Review, deployment and ongoing operations

Areas of expertise

Implementation experience applied to your company's challenges. Explore technical decisions and the scope you can engage me for.

Documents, processing equipment, two model versions and an application; the red route passes through the selected version.

Prepared data, a versioned model and a connected application.

DATA / MODELS / OPERATIONS

AI platforms

Connect data, training and inference with version control, budgets and recovery.

  • From datasets to APIs
  • Training with limits and recovery
A source collection inside a defined space; only one output reaches the review tray.

Protected sources, limited access and an output subject to review.

ACCESS / EVIDENCE / AUDIT

AI with sensitive data

Permissions, protected storage and verifiable output for workflows with human review.

  • Backend authorisation
  • Source checks enforced by code
An application and phone send requests to a queue; the route ends at the confirmation control.

Existing systems, staged requests and confirmation before action.

INTEGRATION / CONTINUITY

AI in existing systems

Integrate with databases, applications and telephony in stages that preserve operations.

  • APIs, events and existing databases
  • Speech and telephony
Three inputs converge into a queue feeding one accelerator board.

The queue organizes models sharing the same GPU.

GPU / ORCHESTRATION

Shared GPU

Coordinate GPU use across text and speech models.

  • Coordinated model use
  • Capacity and waiting-time assessment
An application outside the perimeter accesses a hosted model through a defined gateway.

The application accesses the model on infrastructure defined for the project.

vLLM connected to a product, with memory and concurrency limits.

  • Capacity planning
  • Access control and operations
A microphone, three audio sections and three transcript blocks in the same sequence.

Audio and text follow the sequence of the same conversation.

WEBRTC / OPUS / WHISPER

Speech operations

Real-time audio, transcription and processing recovery from browser to server.

  • Capture by participant
  • Durable audio and queue recovery

Consulting for every stage of operations

From an architecture assessment to ongoing operations. Scope and pricing are defined for your product, data and team.

An architecture plan with a narrow passage highlighted, beside equipment and a measuring tool.

Map the architecture and bottlenecks before defining the deployment.

Infrastructure review

Map data and integrations, identify bottlenecks, assess capacity and plan deployment.

Quoted per project

Request a proposal
A rack and application connected through a panel; the selected integration appears in red.

Connect infrastructure to the application and check the integration.

Deployment

A data and model platform, private inference, speech and integration with your systems.

Quoted per project

Request a proposal
A server, monitoring console and separate copy; a red route indicates recovery.

Monitor the service and keep a defined recovery path.

Ongoing operations

Monitoring, updates, recovery, capacity planning and cost control.

Monthly engagement, quoted

Request a proposal

Concept illustrations generated with AI.

From assessment to operations with your team

Scope, responsibilities and acceptance criteria agreed before deployment.

  1. Assess the workload

    Product, data, volume, latency and budget.

  2. Measure and deploy

    Validate the real workload, integrate and document operations.

  3. Maintain operations

    Monitor failures and costs, tune capacity and rehearse recovery.

Product requests pass through a test bench with a measuring instrument and a criteria sheet.

Product workload guides measurement, capacity and acceptance criteria.

Concept illustration generated with AI.

Reading and experience

Experience

Wendelmaques

12+ years in engineering

Software and telecom. Experience in AI platforms, sensitive data, real-time audio and integration with existing systems.

Consulting

Let's assess your project

Describe your product, current infrastructure and challenge. I will follow up to assess the scope and prepare a proposal.

Your message goes directly to my inbox, without newsletter enrolment.