Implementing AI into ASP.NET.

[Alex_197]

Hi all.

I have a question - how to implement the AI into ASP.NET website?

Long story short.


We have an ASP.NET website based on the MS SQL Server. What we need is to implement the AI to generate summarizing reports based on some tables. Let say Nurses notes, patient vital signs and so on. All the magic should be done locally due to HIPAA restrictions so we can't use cloud services, Azure and something like this.

The idea behind it - the doctor will use our website to see what is going on with his patient for the last month, for example. And the AI should present the summary to the doctor.

It's a completely new area to me and I need your advice on where to start.

Google AI recommends ML.NET but it's very expensive - $275 000.

Any ideas where to start?

Thank you

 

[JohnC]

Here is what ChatGPT says:

Their Goal​

Build an AI-powered summarization system (running on-premises, no cloud) that:

• Analyzes structured and semi-structured data (e.g., nurse notes, vital signs)
• Outputs human-readable summaries of a patient’s recent history (e.g., last month)
• Runs entirely locally, to stay compliant with HIPAA
  

---

What ML.NET Can Do​

ML.NET is good at:

• Predictive modeling (classification, regression, clustering)
• Structured data analysis (numerical and categorical input)
• Integrating into existing .NET + SQL Server stack
• Running entirely on-prem with no internet/cloud requirement

---

What ML.NET Cannot Do Easily​

ML.NET is not designed for:

• Natural Language Summarization (like GPT-style summarizing)
• Understanding unstructured clinical text like nurse notes
• Generating human-like summaries or reports

You’d need:

• A large language model (LLM)
• Or custom NLP models (e.g., using ONNX, Hugging Face transformers) embedded locally

---

The "$275,000" Confusion​

That’s probably a misunderstanding:

• ML.NET is free and open-source.
• There’s no license fee to use ML.NET.
• Microsoft doesn’t charge for it unless you're engaging consulting services.

So, that cost figure might refer to:

• An enterprise AI consulting contract
• Custom development quotes from vendors
• A confusion with Azure AI pricing

---

Where to Start (Realistically)​

1. Define the input​

• Tables: vital signs (structured), nurse notes (free text)
  

2. Choose tools that support on-prem summarization​

If the summaries involve text understanding, look into:

• LLMs running locally:
  • Ollama with models like mistral, llama2, or phi
  • LM Studio – runs GGUF models on local PC
  • [Hugging Face Transformers + ONNX Runtime] in .NET
• Load models like medAlpaca, Clinical-T5, or BioGPT, which are tuned for medical language
  

3. Integrate with .NET / ASP.NET​

• Build a .NET service to:
  • Pull relevant data from SQL Server
  • Format the inputs (structured + unstructured)
  • Feed into a local LLM
  • Display the summary to the doctor in the ASP.NET UI

---

Prototype Suggestion​

1. Start with simple extraction:
   • Show raw nurse notes + vitals side-by-side
2. Use a small local LLM (e.g., Mistral-7B or LLaMA-3) to generate a summary from that text, entirely offline.
3. Later: build a pipeline that:
   • Pulls patient data
   • Feeds it into the local model
   • Returns structured summaries
     

---

Tools Stack Suggestion​

Task	Tool
.NET integration	ASP.NET + ML.NET (optional for data prep)
Model inference (LLM)	Ollama + Mistral / LLaMA 3
Data pipeline	C# backend + SQL Server queries
Summarization Model	Local BioGPT, Clinical-T5, or fine-tuned Mistral
HIPAA Compliance	Keep everything on-prem, no internet access to models

---

Bottom Line​

ML.NET alone is not enough for summarizing nurse notes. You’ll need a local language model that can understand and generate human-like summaries. That’s absolutely doable today, entirely offline, without spending hundreds of thousands.