Loading…
Loading…
Practical AI engineering for developers building real-world systems with .NET, RAG, AI agents, LLMs, SQL Server, and production infrastructure — documented version by version, as it's actually built, not rewritten after the fact.
Every article ties back to a working implementation — not an explainer rewritten from someone else's docs.
ASP.NET Core, C#, LLM APIs, Ollama, Semantic Kernel, streaming responses via SignalR.
Explore →§02Embeddings, chunking, Qdrant, hybrid search, reranking, and production retrieval architecture.
Explore →§03Tool calling, memory, planning, multi-agent orchestration, security and evaluation.
Explore →§04Model selection, prompting, context windows, structured output, cost and latency tradeoffs.
Explore →§05Text-to-SQL, schema discovery, query validation, and giving an LLM safe database access.
Explore →§06Security, observability, evaluation, cost control, and deployment onto real infrastructure.
Explore →Ask a database a question in plain English and get back a validated, schema-aware SQL answer — built and documented version by version, from a simple Text-to-SQL prototype to a secured, RAG-augmented, tool-calling agent.
Every version ships as a real article: architecture, implementation, what broke, and how it was fixed. No step is skipped for the sake of a clean narrative.
Build roadmap
One layer, one job. Nothing is added before the phase that needs it actually ships.
| Layer | Choice | Why |
|---|---|---|
| Public site | Next.js + TypeScript + Tailwind + MDX | Static-first, SEO-friendly, zero database for content. |
| Flagship backend | ASP.NET Core (C#) | Hosts the AI Database Agent — Text-to-SQL, RAG, agents. |
| Accounts & billing | ASP.NET Core + SQL Server | One relational store for identity, tokens, sessions, subscriptions. |
| Vector search | Qdrant | RAG for articles and the database agent. |
| LLMs | Ollama (local) + provider-agnostic API | Keeps the brand independent of any single model vendor. |
| Reverse proxy / TLS | Nginx + Let's Encrypt | Standard, transparent VPS pattern — no managed host required. |
| CI/CD | GitHub Actions → SSH deploy | Push to main, service restarts — no platform lock-in. |
SQL Server and Qdrant never face the internet — only the API is allowed to speak to them.
The connection profile currently wired to this build — Windows-integrated, encrypted, developer machine only. Production swaps this for a scoped service account behind the internal Docker network.
Run a local LLM with Ollama and connect it to an ASP.NET Core application.
Read article →A practical guide to building a Retrieval-Augmented Generation pipeline with .NET.
Read article →Understand Retrieval-Augmented Generation from an engineering perspective.
Read article →