Gael Tuczapski

.NET backend developer, I design business applications and integrate LLMs properly: structured outputs, anti-hallucination guardrails, multi-provider, so that it holds up in production, not just in POC.

Concretely, I solve the problem end-to-end: a .NET backend that handles the load, an LLM component that doesn't drift, and the deployment to production that goes with it.

My approach to AI can be summed up in one sentence: the LLM suggests, the code disposes. Language to the model, never business invariants.

Two examples designed alone: SignalLab-extractor, a pipeline that extracts structured data (schema-constrained JSON) from video content via LLM (Claude, multi-provider Ollama / OpenAI-compatible endpoints, Message Batches API), then verifies it in backtest. And Polaris, a business DMS in a regulated environment (.NET 10, Blazor Server, EF Core 10, tamper-proof audit by SHA-256 chaining, RBAC).

In the field: three years of work-study at Capgemini Engineering at the CEA Marcoule nuclear center, where I single-handedly led the redesign of a business application with ~5,000 users (.NET 8, Blazor Server, Clean Architecture, CQRS, Active Directory SSO). And a team SaaS, Buy&Rent, where I built the data services (collection, processing, and storage from 8 public sources) on a .NET 10 microservices / Kubernetes architecture.

What I'm good at:

- Integrating LLMs into business apps (structured outputs, anti-hallucination, multi-provider)

- Robust .NET backend: ASP.NET Core, EF Core, PostgreSQL / SQL Server, microservices

- Clean and testable architecture: Clean Architecture, CQRS

- From code to deployment: Docker, Kubernetes (K3s), CI/CD GitOps

IMT Mines Alès Engineer. A .NET application to upgrade, an LLM to integrate without it hallucinating, a product to build? Let's talk about it.