Blog · Insights

What we're seeing in AI and data.

Short, jargon-free analysis on LLMs, AI agents, automation, and enterprise data, written by the Nyxora team from the field, project by project.

A hand stops a row of falling dominoes mid-fall.

AI agents in the enterprise: where to start?

Everyone's talking about autonomous agents, but few companies know where to begin. Here's the framework we use with our clients to identify a realistic first use case.

Read the article →
A humanoid robot sitting on a wooden bench, leaning over a brochure on its lap.

How to choose the right language model for your project

Quality on your task, latency, cost, confidentiality, vendor lock-in: the real selection criteria, and why you should keep the ability to switch.

Read the article →
Close-up of a handshake between two people in shirts, in an office.

Case study: an AI agent that prepares your sales meetings

Prospect research, meeting brief, CRM updates approved by the salesperson: a useful agent, and what it must never do.

Read the article →
A bright desk with a laptop and a monitor displaying “Work hard anywhere”.

Working with a remote AI team: how we protect your data

Contracts, access, data handling, end of engagement: the concrete rules we apply on every engagement.

Read the article →
A laptop on a glass table displays a dashboard with charts and indicators.

Automated reporting: dashboards that comment themselves

Connect the sources, compute in code, let AI write the commentary and keep a human review.

Read the article →
A person gesturing with their hands while explaining something in a meeting, in front of a laptop and a notebook.

Getting teams on board with AI: change management that works

Understand the obstacles, involve users, train on real tasks, rely on champions: how to get an AI tool adopted for the long term.

Read the article →
A hand holds a magnifying glass over printed account statements.

AI audit: what we actually look at in the first two weeks

Before any project, we map the terrain. What a real AI audit inspects, and why it matters.

Read the article →
White and cream envelopes of various sizes laid flat, alongside paper tags.

Automating the sorting and replies of incoming emails

Classify, extract, route, draft replies: how to automate a shared mailbox without losing control of what gets sent.

Read the article →
A dark wooden chess pawn among light wooden pieces on a chessboard.

Build in-house or outsource your AI skills?

In-house team, provider, staff augmentation or hybrid: the criteria for choosing, and the conditions for staying in control.

Read the article →
A long library aisle lined with shelves full of books, lit by hanging light bulbs.

RAG in practice: building a reliable document assistant

Documents, chunking, access rights, hybrid search, citations, evaluation: what separates a RAG prototype from a reliable assistant.

Read the article →
A hand holding a pen over a handwritten checklist in a squared notebook.

Automation: the 5 processes an SME should automate first

Four criteria, a simple scoring method and five processes that make excellent first projects.

Read the article →
A metal padlock resting on a laptop keyboard, lit in red and green.

Is your data going to OpenAI? What really happens when you send a document to an LLM

Retention, training, hosting: what really happens to your data, and the habits that keep you in control.

Read the article →
A hand holding a pen signs a paper document on a table.

Automating supplier invoice processing, step by step

Receipt, extraction, checks, approval, booking: where to use AI, where to leave it out, and how to stay in control.

Read the article →
A hand points at a laptop screen while another person types on the keyboard.

Testing an AI agent before (and after) go-live

Real scenarios, outcome and path evaluation, regression testing, production monitoring: how to test an agent properly.

Read the article →
Abstract visualization of a neural network with orange accents.

No, an LLM doesn’t “understand” what you tell it

What actually happens under the hood, and why it changes how you should use it.

Read the article →
A handwritten monthly planner open on a white table, with a cup of coffee resting on it.

Building a 12-month AI roadmap

Inventory, a first measurable project, governance, scaling up: a realistic AI roadmap, quarter by quarter.

Read the article →
A young green sprout growing out of a pile of coins.

What does an AI project really cost?

The cost items beyond development, and a simple method to estimate return on investment.

Read the article →
Two people talking in front of their laptops, black and white photo taken through a glass pane.

Human in the loop: designing agents that know when to stop

Approval points, escalation, confidence thresholds, traceability: how to design an agent that knows when to ask a person.

Read the article →
Bronze statuette of blindfolded Lady Justice holding scales.

GDPR and AI: what we check before sending any data to a model

Legal basis, minimization, processors, transfers outside the EU, DPIA: our GDPR checklist before any AI project.

Read the article →
Black and white close-up of an interlocking gear mechanism.

Automation: the mistakes that sink a project

Technology is almost never the real problem. What actually makes an automation project fail, and how to avoid it.

Read the article →
An open laptop on a long, bright desk facing a large window, next to a green plant.

Automating a new employee’s arrival

Accounts, equipment, documents, welcome path, internal assistant and departure: automating onboarding without losing the human welcome.

Read the article →
A hand works on an open electronic circuit with a precision screwdriver.

From POC to production: why so many AI projects never get past the prototype

Demo data, no integration, no owner: why AI prototypes get stuck, and how to design a POC that makes it to production.

Read the article →
A half-open laptop in the dark, its screen casting an orange and blue glow.

Shadow AI: your teams already use ChatGPT, do you know it?

Why shadow AI exists, what it really exposes, and how to govern it without banning it.

Read the article →
A fountain pen resting on an open notebook with “Notes” written on it, a pair of glasses in the background.

Writing good prompts in a business: method and common mistakes

Context, task, constraints, examples, output format, versioning and testing: writing reliable prompts rather than magic formulas.

Read the article →
A laptop in the dark, its screen covered in columns of green characters.

Prompt injection: the flaw every AI project must plan for

Why an LLM can be manipulated, why agents and RAG are on the front line, and the safeguards that really limit the damage.

Read the article →
Close-up of a numbered network patch panel with blue and grey cables plugged in.

Connecting an AI agent to your tools: CRM, ERP, email

APIs, MCP, service accounts, least privilege: how to plug an agent into your business tools without handing it the keys to the whole company.

Read the article →
Close-up of a neatly stacked pile of white paper sheets.

RAG or fine-tuning: how to choose?

Two approaches, two costs, two very different use cases. A simple guide to deciding based on your needs.

Read the article →
A system of equations printed on a page, partly blurred and partly sharp.

Hallucinations: why they happen and how to reduce them in production

Why an LLM confidently makes things up, and the concrete levers that make an AI system reliable.

Read the article →
A red-marker flowchart drawn on a whiteboard.

How to choose your first AI use case

The best AI project isn't the most impressive one, it's the most measurable one. Our 4-criteria selection grid.

Read the article →
A small child at the foot of a large sunlit stone staircase.

The AI Act: what SMEs really need to know

Risk levels, the deployer role, transparency, staff training: what the European regulation concretely changes for you.

Read the article →
Server racks in dim light, threaded with orange and blue cables and dotted with green status lights.

Open models: when hosting your own model makes sense

Confidentiality, control, offline use, volumes: when to host an open model, what it involves, and the middle-ground options.

Read the article →
A grid of yellow sticky notes on a wall, a hand placing the last one.

AI agent or simple workflow: what do you really need?

Not everything needs an autonomous agent. Our decision grid for choosing between an agent and a workflow.

Read the article →

A topic worth exploring together?

If one of these topics resonates with a project you're running, let's talk about it during a free consultation.