Experiments with generative AI applied to public sector problem statements.
The Defense Threat Reduction Agency (DTRA) issued a SBIR topic looking for solutions to analyze scientific articles to detect potential proliferation-related threats. This is a problem that could benefit from a multimodal approach, with the model having to analyze text, images, and tables in order to detect threats. Mindflayer implements several processes to deal with the multimodal aspect of the problem, and includes a demonstration of concept to process a PDF and extract relevant information, labeling it, and storing the results in different structured formats.
Market research and data analysis efforts for AFWERX 24.7 Open Topic SBIR. The artifacts in this folder present a portion of the work done by the author for the SBIR Phase I proposal.
A series of notebooks that generate benefits/disability claims data based on VA Form 21-526EZ for zCore Group's NWQ project (ongoing). Raymond Gonzalez, research partner.
Another company RAG (WIP). The first notebook parses and cleans meetings transcripts (from Supernormal but any transcription content can be used). Tokenizer included. Qdrant or Chroma can be used.
© 2025, Michael Nau. All rights reserved. Unauthorized reproduction, distribution, or use of this material is prohibited.