--- title: "AI integration for your .NET applications | GiLabs" description: "AI built into your .NET applications: RAG on SQL Server, Semantic Kernel and Agent Framework agents, document extraction, in your code and your IT system." url: "https://gilabs.fr/en/services/dotnet-ai-integration/" lang: "en" service: ".NET AI integration" --- # AI integration for your .NET applications: RAG, agents and Semantic Kernel inside your IT system *Service · deep integration* For mid-caps and software vendors whose IT lives in .NET and SQL Server: AI developed inside your application (document extraction, agents, classification) with your business rules, your architecture, your code. Not another tool next to the IT system: a new capability inside it. I work in your repo, alongside your developers, to your team's standards. **Book a 30-minute call**: [Calendly](https://calendly.com/yann-gilabs/30min) ## This is for you if… - Your IT system or product rests on a .NET / SQL Server estate that AI should enrich, not bypass. - You're aiming for a native AI capability: document extraction, classification, business assistants, directly inside the application your users already know. - Your context is demanding: industry, banking, healthcare, regulated environments, where traceability and GDPR / AI Act compliance are non-negotiable. - You have a technical team that wants a counterpart who can talk architecture, not just prompts. ## What I deliver - **Document extraction in production**: Scans and inbound documents turned into structured data: a swappable extraction engine (Azure Document Intelligence, Mistral OCR, Docling), business rules, transactional import. - **RAG on your SQL Server data**: Your documents and data queryable from the application, with sourced answers: a vector index in SQL Server (VECTOR type) or Azure AI Search next to the database, chunking that follows the business structure, IT system permissions respected. - **Agents integrated into the IT system**: Semantic Kernel and Microsoft Agent Framework, on Azure OpenAI or another provider: agents that query your data and act inside your application, with your IT system's permissions and traceability. - **Clean architecture**: Clean Architecture, DDD, tests: code your team can read, maintain and extend. Delivered in your repo, under your CI/CD. - **Full reversibility**: No dependency on GiLabs: documentation, knowledge transfer, and abstractions that let you switch AI providers without a rewrite. ## The process The method of a serious software project, applied to AI, because that's what it is. 1. **Architecture review**: Your IT system, your flows, your security and compliance constraints. We decide together where AI belongs and where it has no business being. (Deliverable: an architecture dossier validated by your technical team.) 2. **Iterative development**: Short sprints on your real data, code reviews with your team, quality measured at every iteration. (Deliverable: tested, deployable increments, in your repo.) 3. **Production and handover**: Deployment into your infrastructure, monitoring, documentation, training for the team taking over. (Deliverable: the AI capability in production, mastered by your developers.) ## Investment This is the most committing tier, and the one that creates a durable asset: an AI capability your team owns, in code it masters, with no rent to pay to yet another vendor. Investment: Fixed fee. Scoped in the architecture review, depending on integration depth. Scope, price and acceptance criteria agreed before the first line of code. ## Frequently asked questions ### Our developers are good. Why not build it in-house? Often you can, and that is the right answer once the first project is behind you. What is rarely missing is .NET skill; what is missing is having already run an AI chain in production and knowing where it breaks: thresholds, error recovery, model drift, cost per document. The most common tell is a company hiring a .NET developer and a data scientist at the same time, when it is looking for one profile. I work in your repository, with your team, and the skills transfer is part of the fixed fee. ### How do you work with our technical team? In your repository, under your CI/CD, to your standards, with code review both ways. Clean Architecture, DDD and tests are not sales arguments here: they are what makes the code readable by your developers. The project starts with an architecture review that decides with them where AI belongs, and above all where it does not. ### Does our data leave the company, and leave Europe? That is an architecture decision, settled before the first line of code, not a side effect of picking a model. Depending on your constraints, processing stays on your own infrastructure, on EU-hosted services, or on US services under contractual safeguards. For regulated environments and professions bound by professional secrecy, the default is strict: processing in Europe or on premises. The traceability required by GDPR and the AI Act is designed in, not retrofitted. ### Will we be locked into one model provider? No, and that is an explicit architecture choice. Extraction and inference components sit behind abstractions: moving from Azure Document Intelligence to Mistral OCR or Docling, or from one model provider to another, happens without rewriting your business rules. That is what makes the capability last in a field where the state of the art moves every six months. ### Semantic Kernel or Microsoft Agent Framework? Microsoft presents Agent Framework as the successor to Semantic Kernel for agents. Semantic Kernel code in production keeps working: the real question is when to migrate, not whether to rewrite everything. My default rule: new agents start on Agent Framework, existing Semantic Kernel code migrates when a piece of work touches it, never in one go. The architecture review decides based on your estate. ### Do we need a new database to do RAG? No. The source of truth for your data and permissions already lives in your SQL Server: the vector index is added there (SQL Server 2025 and Azure SQL have a VECTOR type), or lives next to it in Azure AI Search when volume or hybrid search justify it. Either way, no diverging copy of your data in yet another database. ## On the same topic - [Integrating AI into an existing .NET application: the architecture guide](https://gilabs.fr/en/guides/integrer-ia-application-dotnet/): Four ways to wire AI into a .NET application already in production, where the code belongs, and the guardrails that hold: typed output, deterministic checks, tools under the user's identity, writes behind approval. Tested C# code on GitHub. - [Semantic Kernel or Microsoft Agent Framework: what to do with an existing .NET application](https://gilabs.fr/en/guides/semantic-kernel-agent-framework/): Agent Framework 1.0 shipped in April 2026 and Semantic Kernel has moved to maintenance. What changes in a .NET application, what stays put, the code before and after, and when to migrate. - [RAG on SQL Server 2025: a reference architecture for an existing .NET application](https://gilabs.fr/en/guides/rag-sql-server-2025/): Wiring RAG into a .NET and SQL Server application without a separate vector store: the VECTOR type, chunking that follows the documents, permissions enforced in the query, an Agent Framework agent. Complete, tested code on GitHub. - [The AI POC that will never reach production as-is](https://gilabs.fr/en/blog/industrialiser-poc-python-dotnet/): A Python POC that convinces in a demo but stays stuck for six months does not need to be rewritten in C#. It needs to be exposed behind an API and brought into the rest of the estate's rules. - [Why I chose Semantic Kernel over Python for AI in .NET](https://gilabs.fr/en/blog/certification-az2005-semantic-kernel/): Holder of the Microsoft Applied Skills AZ-2005 credential, I chose to integrate AI into existing .NET systems rather than rewrite everything in Python. The reasoning, and the Semantic Kernel pattern. ## Going further - [Case: document AI pipeline](https://gilabs.fr/en/work/document-ai-pipeline/): Exactly this kind of integration, in daily industrial production. - [30-day AI diagnostic](https://gilabs.fr/en/services/ai-diagnostic/): Not sure about the scope? The diagnostic quantifies and prioritises before you commit to the fee. - [AI agents and business applications](https://gilabs.fr/en/services/ai-agents-automation/): No .NET estate to build on? Agents and applications are also built on Microsoft 365 and the tools you use every day. **Book a 30-minute call**: [Calendly](https://calendly.com/yann-gilabs/30min)