AI Automation & Agent Systems

Engagement options

• AI Discovery Workshop (1-2 weeks): from EUR 2,500
• Pilot agent or workflow build (6-12 weeks): from EUR 8,000
• AI Ops retainer: EUR 4,500 per month

Frequently asked questions

What does an AI automation engagement with Matthieu Pesesse look like?

Engagements typically begin with an AI Discovery Workshop (1-2 weeks) to map opportunities and define success metrics, followed by a pilot agent or workflow (6-12 weeks) deployed on production infrastructure. Long-term retainers cover orchestration, prompt iteration, and reliability monitoring.

How much does AI automation cost?

An AI Discovery Workshop starts at EUR 2,500. A focused pilot agent build typically runs EUR 8,000-15,000. Ongoing AI ops retainers start at EUR 4,500/month. Final pricing depends on integrations, model choice, and reliability requirements.

Which LLM providers and frameworks does Matthieu Pesesse use?

Matthieu Pesesse deploys on OpenAI GPT-class models, Anthropic Claude, NVIDIA NIM for on-premises GPU inference, OpenClaw for multi-agent orchestration, and n8n for workflow automation. Selection is driven by latency, cost, data-residency, and capability requirements rather than vendor preference.

Can the AI run on-premises or in EU data centres?

Yes. Matthieu Pesesse supports fully on-premises deployment via NVIDIA NIM and open-weight models (Llama, Mistral, custom fine-tunes), as well as EU-hosted cloud inference for organizations with GDPR or sector-specific data-residency requirements.

What types of workflows are good candidates for AI automation?

High-volume, rules-light, decision-bounded processes — document classification and extraction, customer support triage, content generation pipelines, internal Q&A on private corpora, code-assistance agents, and multi-step research tasks. Hard-rule deterministic processes are usually better automated without LLMs.

How is reliability monitored once an AI agent is in production?

Each engagement ships with structured observability: input/output logging, eval harnesses on golden datasets, drift detection, and human-in-the-loop escalation paths for low-confidence outputs. Monthly reviews adjust prompts, model selection, and retrieval pipelines.