Incident Copilot
In development
Context
A project in development for exploring a small web application that helps organize synthetic technical-incident evidence and returns a structured response for human review.
Problem
Incident evidence can arrive scattered across a title, symptoms, context, and logs. The objective is to study a bounded integration that helps formulate hypotheses without inventing a root cause or replacing technical validation.
Objective
I am defining an experimental flow with validated input, preventive secret sanitization, a call to an LLM provider, and validation of a structured response.
Technical decisions
- I am using C# and ASP.NET Core for a small web application focused on one use case.
- I am sanitizing input before sending it to an external provider and working only with synthetic data.
- Keep structured responses and controlled messages for configuration errors, limits, timeouts, or invalid responses.
- Treat the LLM integration as a bounded exploration and a support tool for human review.
Planned scope
- The planned scope includes a form, validation, sanitization, assisted analysis, and presentation of a structured response.
- The LLM integration is not presented here as a finished capability.
Planned architecture
The planned flow connects a form, validation, preventive redaction, an LLM provider, JSON parsing, result validation, and a review view.
Privacy and limits
The project must work with synthetic incidents. Redaction reduces accidental exposure, but it does not guarantee safe input or remove the need for human review.
Current status
The project is still in development. The LLM integration and capabilities described as scope are not presented as a finished product.
Limits
- The project is still in development and depends on external configuration for a real analysis.
- Sanitization is preventive and cannot guarantee detection of every secret; synthetic data must be used.
- It is not presented as a production-ready solution.
Current status
In development. The case explores a bounded and responsible LLM integration, but the real integration is not presented as finished or production-ready.
What it will demonstrate
- Validation and privacy should exist before evidence is sent to an external provider.
- A structured response supports human review but does not replace technical diagnosis.
Technologies
- C#
- .NET
- ASP.NET Core
- LLM API
- HttpClient
- System.Text.Json
- xUnit