// Service
AI Systems & Agents in Production
LLM applications, RAG, agentic workflows, fine-tuning, evaluations — built like systems, not research projects.
Book a call →This is our flagship practice in 2026. The thesis is simple: most AI projects don’t fail because of the model. They fail because nobody treated them like a real system.
What we ship
AI systems that actually run in production — with controlled costs, real observability, and a plan for when they break. Not a demo that almost works, not a POC pushed “as-is” only to be found unmaintainable a few weeks later.
Our approach always starts with the same question: does this problem deserve an LLM, or would a 200-line script do the job? When AI is the right answer, we build everything around the model that makes it operable: evaluations, observability, guardrails, cost model.
What it covers
- Agentic systems — tool-use, state management, multi-step orchestration, business guardrails
- RAG & retrieval — vector store choice, hybrid search, domain-aware chunking, evaluation strategies
- Fine-tuning — SFT, LoRA, dataset curation, evaluation for low-resource or domain-specific problems
- LLM Ops — cost model, latency budgets, eval harnesses, per-call observability, rollback
- Existing AI system audit — diagnostic on a project stuck between POC and production
Where we make a difference
Most AI providers ship a notebook that works. We ship a system that survives on-call. Concretely: we define evals with you that catch regressions before they reach users, we instrument every model call so a case can be debugged three weeks later, and we define the rollback procedure before the first production push — not after the first incident.
Typical engagement formats
- Diagnostic (1 week) — assessment of an AI project stuck between POC and production, with actionable recommendations
- Build engagement (2–4 months) — end-to-end design and delivery of an AI system, co-built with your team
- Advisory — architecture review, second opinion on a technical decision, pre-launch support
Currently shipping
- SAAGA Solve — agentic AI as a digital workforce
- Campalingua — fine-tuned LLMs running in production for Corsican language