Equipping a tracking platform for AI: Secure, local RAG prototype with .NET and Semantic Kernel

Services included:

Architecture
Software development
Custom AI solution
Technical research & feasibility

A developer of airborne object tracking platforms turned to Resolute for help building a modern, secure AI prototype. After a successful collaboration on both web and desktop real-time tracking systems, the client wanted to explore how artificial intelligence could support users' queries and reference complex operational regulations.

Resolute delivered a fully localized Retrieval-Augmented Generation (RAG) proof of concept, capable of answering questions from thousands of pages of regulations with traceable accuracy. Entirely built in .NET using Microsoft’s Semantic Kernel, the solution strikes a balance between performance, control, and extensibility, with no reliance on third-party models or external cloud APIs.

Challenge

The client’s operational domain involved large volumes of legally binding documentation - aviation and defense compliance, safety regulations, and other institutional standards. The request: build an AI assistant that could parse these documents and return specific, verifiable answers.

The constraints were significant:

Vast volumes of source material - around 100 documents, each exceeding 1,000 pages, in PDF and XML formats
Limited internal understanding of AI capabilities
Modest infrastructure, no dedicated GPU cluster
Demand for full traceability and citation of results

Solution

Resolute proposed and built a complete RAG (Retrieval-Augmented Generation) solution, hosted and run locally using lightweight open-source components. It was tailored for both the client's infrastructure and the specific nature of their documents.

The AI assistant follows a retrieval and response workflow designed to handle large, unstructured documents and generate accurate, reference-based answers.