AI development_

AI development is mostly engineering

A language model is a component, like a database. The work that decides whether an AI feature succeeds is everything around it: getting the right data in, keeping it fresh, controlling who sees what, catching bad answers, and keeping the system fast and affordable when real users arrive. That is software engineering, and it is what we have been doing for broadcasters, manufacturers and restaurant chains.

Instead of a slide of AI logos, here is the engineering record that AI work depends on.

Live data from many sources

For N12 we built the Corona N12 dashboard: custom-made with .NET, pulling and presenting live data from multiple sources, including the Israeli health ministry and global health sources. Merging, cleaning and refreshing data from sources you do not control is exactly what a retrieval pipeline does before an assistant can answer anything.

Real-time results under peak load

For Mako we built the elections site that shows Israel’s election results in real time. It is probably the most popular site for following the results, and we have run it for more than one election. Spikes in traffic and zero tolerance for wrong numbers are a good rehearsal for putting AI in front of the public.

Operational data turned into decisions

For Netafim we built a web-based OEE application that gives real-time insight into production performance. Turning raw factory data into something a manager can act on is the same skill as turning raw documents into answers.

Platforms that run a business

ibon is a management platform for the food industry, with real-time inventory, production lifecycle and regulatory compliance, used in over 80 Aroma branches. Business-critical systems with many users and strict rules are where AI features have to earn trust.

Choosing the right approach

Not every problem needs an agent. A quick guide to what we usually recommend:

You want to…	What we build
Answer questions from your own documents	RAG assistant with citations
Handle customer messages on several channels	GPT assistant connected to your CRM and inbox
Run a multi-step process across tools	AI agent with human approval on risky steps
Add AI to a product you already sell	AI features inside your codebase
Feed AI and dashboards with clean data	Data pipeline, scheduled or real time

Often the honest answer is a mix: a pipeline that keeps your knowledge base fresh, a RAG assistant on top, and one or two automations that act on what it finds.

Accurate, private and affordable

Three things decide whether an AI feature survives its first month in production.

Accuracy. We build a test set from your real cases before we tune anything, and every change is measured against it. Answers are grounded in your sources and cite them.

Privacy. Permissions follow the user, sensitive fields can be masked before they reach a model, and every call is logged. Where data residency matters, hosting and providers are chosen to match.

Cost. Model choice, caching and prompt size all change the running bill. We track cost per conversation or per task from day one, so there are no surprises when usage grows.

Where AI development fits

If you want AI across the whole organisation (use-case selection, privacy, training, change management), start with AI implementation. If the AI needs to sit inside your sales or service process, see custom CRM development, where we build assistants that answer leads by SMS, chat and email. For the wider product around the AI, see custom software development, and for a practical roadmap read how to implement AI in your organisation.

Bring us one job you would like AI to do, and we will tell you how we would build it and how we would prove it works.