Kitsoft Case Study: Generative AI for GovTech

Ingram Technologies partnered with Kitsoft, the GovTech company behind Ukraine's Diia platform, to bring generative AI to BelDoc — a fully digital service for incorporating a company in Belgium. We built an AI assistant that guides founders through registration, reviews the legally mandatory financial plan, and suggests the right NACE-BEL activity codes from a plain-language business description. The result turns one of the most paperwork-heavy moments in a founder's life into a guided, 20-minute conversation.

The Partner and the Problem

Kitsoft has spent more than a decade digitalizing public services. Their work spans 200+ public services and 22 million users, including Diia, Ukraine's flagship digital-state platform, and their open-source low-code platform Liquio. With offices in Kyiv, Brussels, and Berlin, they bring deep GovTech expertise to the European market.

BelDoc is their answer to a stubborn problem: incorporating a limited liability company in Belgium (an SRL/BV) is slow, fragmented, and intimidating. Founders juggle legal templates, notary appointments, a financial plan, VAT registration, and economic-activity classification — usually across several platforms and several weeks. BelDoc consolidates all of it into a single guided flow, with notarization handled online and documents delivered to a secure wallet.

The flow was already excellent. What it lacked was judgement — the kind of guidance a founder would otherwise pay an accountant or lawyer for. That is where Kitsoft asked us to help, and where generative AI fits GovTech precisely: not replacing the legal process, but making expert reasoning available to every applicant, in their own words, at the moment they need it.

What We Built

We designed and implemented a suite of generative-AI features that sit inside the BelDoc registration flow. Each one targets a specific decision point where founders typically get stuck.

An AI Advisor for Company Registration

The centerpiece is a conversational advisor that walks founders through incorporation in plain language. It explains what an SRL/BV is, the difference between share classes, what the founders' contributions imply, and what each step of the notarial process actually requires. Crucially, it is grounded in Belgian company-law context rather than generic answers — it knows the difference between what BelDoc collects and why the law requires it. Founders ask questions the way they would ask a human advisor ("do I need a registered office if I work from home?", "what changes if I have a co-founder abroad?") and get accurate, context-aware answers without leaving the form.

AI Analysis of the Mandatory Financial Plan

Belgian law requires every SRL/BV to file a financial plan with the notary at incorporation. It is not a formality: if the company becomes insolvent within three years, founders can be held personally liable when the plan was manifestly inadequate for the activity undertaken. Most first-time founders have no idea whether their plan would survive that scrutiny.

We built an AI layer that reviews the financial plan as it is drafted. It checks the plan for internal consistency, flags assumptions that look unrealistic for the stated activity, identifies missing elements that notaries expect to see, and explains — in clear terms — where the founder's projections are thin. The goal is not to produce a rubber-stamp document, but to raise the quality of the plan and reduce the legal exposure that founders rarely understand they are taking on.

AI Suggestion of NACE-BEL Codes

Every Belgian company must declare its economic activities using NACE-BEL codes when registering with the Crossroads Bank for Enterprises. The official nomenclature runs to hundreds of granular categories, and choosing badly causes real friction later — with VAT, with banks, with sector regulations. Founders routinely guess.

We replaced the guessing with generation. A founder describes their business in a sentence or two — "we run a small online shop selling handmade ceramics and also offer pottery workshops" — and the model proposes the most appropriate NACE-BEL codes, ranked, each with a short rationale for why it fits. The founder confirms or adjusts. This single feature collapses what is normally an opaque lookup into a natural-language interaction, and meaningfully improves the accuracy of the codes that end up on the official record.

How We Approached It

GovTech raises the bar on a few things that consumer AI products can afford to be loose about. We treated three of them as non-negotiable.

Grounding over guessing. Legal and financial guidance cannot hallucinate. We grounded the models in curated, jurisdiction-specific context so that answers reflect Belgian requirements rather than a plausible-sounding average of the internet. Where the model is uncertain, it is built to say so and point the founder to the human notary in the loop.

Human authority stays human. The notary still reviews and approves. The AI accelerates and clarifies the founder's side of the process; it never asserts legal authority it does not have. This division of labour is what makes generative AI appropriate for a regulated process rather than reckless in one.

Privacy and trust by design. Incorporation involves identity documents, ownership structures, and financial projections. We built the AI features to handle that data with the care a public-facing GovTech service demands, in line with the responsible-AI principles that guide all of our work.

Why It Matters

This project is a clean illustration of where generative AI earns its place in government and public-service technology. The hard parts of company formation were never the forms — they were the judgement calls hidden inside them: Is my financial plan good enough? Which activity codes are right? What does this step actually mean for me? Those are exactly the questions that expert reasoning answers, and exactly the kind of reasoning generative models can now make available at scale, instantly, and in the founder's own language.

Working with Kitsoft, we took a process that historically required an accountant, a lawyer, and several weeks, and embedded the relevant expertise directly into the product. For founders, that means fewer mistakes and far less intimidation. For the public registry, it means cleaner, more accurate filings. And for GovTech more broadly, it is a template: meet citizens where they are, ground the AI in the law, and keep the human authority exactly where it belongs.