I ship AI agents, RAG systems, workflows and copilots in production, not in a demo.

Book a call

Capabilities

Six areas of expertise, one goal: production.

From model choice to a system running in production, here's what I take on.

See the capabilities

Use cases

Examples, not a catalog.

Here are typical cases. Yours is unique: that's exactly what our first call is about.

Sectors

Your industry, not just AI.

A good AI project starts by understanding your business, not by picking a model. I have shipped to production across several worlds, same building blocks but very different problems. Yours isn't listed? Write to me, I'll often have a close case.

See the work

Work

AI systems in production.

A few recent projects: MCP infrastructure, agents, RAG, observability. The starting problem, what I built, and what it changes.

  • AI infrastructure · Security

    Give AI agents access to internal tools without opening a hole.

    An MCP infrastructure that decouples LLMs from executing information-system tools: a Zero Trust / SSO proxy in front of every tool, secrets never exposed, and observability on every LLM call.

    Every LLM wired to each internal tool by hand

    One protocol exposes the tools to agents, under access control

    • MCP
    • fastmcp
    • Node.js
    • Zero Trust
    • SSO

  • Anomaly detection · Agents

    Detect capacity anomalies, and have agents explain them.

    A three-engine detection system (statistical, machine learning, hybrid) across multi-OS fleets, with LLM agents orchestrated by LangGraph that explain and prioritize the anomalies. Interactive dashboards and automated PDF reports.

    Capacity thresholds watched by hand, alerts with no context

    Three-engine detection, explained and prioritized by agents

    • LangGraph
    • Python
    • FastAPI
    • Next.js
    • ML

  • RAG · Knowledge base

    A technical knowledge base that answers, always in sync with the code.

    An end-to-end RAG chain feeding the internal AI assistant, and an idempotent synchronization service between the technical GitHub repositories and the knowledge base.

    Docs drifting from the code, answers hunted down by hand

    An assistant answering on a continuously synced base

    • RAG
    • Kibana
    • Python
    • GitHub
See the work

Work done on assignment. Full context on a call.

Method

From audit to production, no detours.

Every project follows the same pipeline. You always know where the system stands, what's done and what's running.

The first steps take days or weeks. The last one never stops.

Audit → POC → Build → Deploy → Run
  1. Auditthe lay of the land, cases prioritized by ROI
  2. POCa prototype that proves the value
  3. Buildthe system shipped to production
  4. Deploycloud or on-prem, monitored
  5. Runevals, observability, continuous optimization

Working together

Four ways to start.

Every system is specific to your case, so there's no off-the-shelf price. The firm quote comes after the audit, once the real scope is clear. No nasty surprises, and we scope the right format on the very first call.

  • 01

    AI Audit

    An honest lay of the land, in a matter of days: where AI would pay off, and where it would just burn money.

    Who it's for

    You're unsure where to start, or you have doubts about a project already underway.

    What you get

    • Your processes put under the microscope
    • Use cases prioritized by ROI
    • Pitfalls and costs flagged early

    Deliverable

    A clear report and a costed roadmap.

    Book a call
  • 02

    POC Sprint

    A prototype that proves the value on your real data, in a few weeks.

    Who it's for

    You want to de-risk before investing in a full build.

    What you get

    • One targeted use case
    • A prototype on your real data
    • An evidence-backed go/no-go decision

    Deliverable

    A working POC and the numbers to make the call.

    Book a call
  • 03

    Production Build

    A reliable, monitored system in production that your teams keep using. A few weeks to a few months depending on scope.

    Who it's for

    The use case is proven; now you need a system that runs for real.

    What you get

    • A deployed system, cloud or on-prem
    • Evals, guardrails, observability
    • Documentation and handover

    Deliverable

    A production-ready product that you own outright.

    Book a call
  • 04

    Retainer & optimization

    I run it, keep it reliable and bring the bill down, month after month.

    Who it's for

    You have AI in production to maintain, improve or cut costs on.

    What you get

    • Ongoing operations and monitoring
    • Cost optimization, 30 to 70% less
    • Enhancements and regression testing

    Deliverable

    A system that stays reliable and profitable.

    Book a call

Do the math before you call me

How much AI could save you, before we even talk.

Three numbers are enough for an estimate. You get a ballpark of the yearly gains, something to put on the table internally to make the case. It's an estimate, not a quote: the price of a system depends on your case, and we scope it on the call.

data entry, document search, email triage, reporting

around 5,850 EUR of time freed per year

that's about 130 hours a year handed back to your team for higher-value work.

A deliberately conservative figure. It counts time only, not the errors avoided or the delays cut, which often weigh more. It also doesn't subtract the setup cost: that's exactly what we scope together.

See if my case holds up

Why guinat

Why trust me with your AI.

  • Production, not demos

    I ship systems that run for real and that your teams keep using, not POCs that gather dust.

  • ROI first

    I only do AI where it pays off. I'll advise against the cases that aren't worth it.

  • Your data stays yours

    Cloud or on-prem, EU hosting, private models: your data stays inside your walls.

  • One point of contact

    I'm Nathan. From audit to deployment, you talk to the person who designs, codes and ships, not a middleman.

  • Costs under control

    Tokens are expensive. Caching and the right model in the right place: I routinely see 30 to 70% savings.

  • Reliable and measured

    Evals, guardrails, observability: systems you can trust, not black boxes.

Resources

What I write about AI in production.

Concrete notes on LLM costs, enterprise RAG and agent reliability.

See all articles

Straight answers

The questions people ask me about AI.

The real worries, answered straight. Still have one? Email me, I reply within 24 hours.

Another question? Email me

Let's talk

Let's talk about your use case.

Book a call[email protected]

Free, no-obligation first call. At the very least you walk away with a straight opinion.