About

What I do

Most of my code is written by AI now. That is a discipline, not a shortcut. I set the architecture and the constraints, hand the mechanical work to coding agents, and review everything before it reaches production. The value was never in the typing. It is in capturing what a business actually needs, and in improving the brief instead of just building it.

To make that reliable, I build the machinery around it: FastMCP servers so agents work from real context instead of guesswork, and a code-search engine written in Rust (DuckDB, BM25, hybrid retrieval) that reads across many repositories at once, so nothing falls between them.

Underneath it all: Svelte, Blazor and Astro on the front end; C# / .NET and Azure behind; Python and pyscript for the automation glue.

If you want someone who treats AI as engineering rather than magic, that is the job.

About D-Soft

D-Soft is a software consultancy. It builds the systems that let software agents do real work — context servers, retrieval engines, web platforms engineered to be read by machines as well as humans, and the agentic-commerce readiness that makes a shop legible to AI shopping agents.

What D-Soft builds

• Agent infrastructure. FastMCP servers that give coding and commerce agents real context instead of guesswork, and Model Context Protocol surfaces that let an agent call real operations.
• Retrieval and code search. A retrieval engine written in Rust (DuckDB, BM25, hybrid retrieval, tree-sitter chunking) that reads across many repositories at once, so nothing falls between them.
• Web platforms. Svelte, Blazor and Astro on the front end; C# / .NET and Azure behind; Python for the automation glue. Static where static is enough, with a thin server edge where it is not.
• Agentic commerce readiness. Making webshops discoverable and transactional for AI shopping agents — structured catalogues, a /.well-known/ucp manifest, and the ordering surface that lets an agent actually place an order.

Proof, not slides

The site you are reading is the reference. It serves an /llms.txt index, ProfessionalService structured data, and a strict security header set, so AI crawlers read it correctly and safely today. The agentic-commerce-readiness offering is built on the same code-search and agent-infrastructure work that makes coding agents reliable here. Eat your own dog food first.