The purpose of this application is to generate a Bill of Materials and a high level networking and interconnect topology diagram and optionally output build configuration sets.
This app uses a supermicro product list to scrape all relevant documentation for configuration options. This is trained against both an LLM (options openai or local) for human interaction and then trained against a recommender engine for configuration recommendation using one shot transformers.
Phase 1 provides a textural prompt quenstion and response only using Riva text processing, Phase 2 will include full conversational interaction leveraging nvidia Riva. This document only contains Phase 1 instructions and guidelines
Note: See INSTALL.md for installation instructions.
Product data is stored in a cuDF dataframe https://github.com/rapidsai/cudf for performance over cpu based pandas.
Prompting is managed by NeMo Guardrails langchain