My main goal here was to write an agent that would use context from Global Security(http://www.globalsecurity.org/) to answer intel-based questions, like simulating the first few days of a conflict between the US and Greenland or Iran.
This took longer than I thought it would because I kept getting distracted with EMT class as well as quantum physics ideas like flipping an entangled qubit. What is Schrödinger’s cat? Is it our measurement of the wave function that is causing it to collapse in the double-slit experiment? As you can probably tell, I am still very much a neonate in that subject. It’s only the beginning, and there will be some future blog posts on quantum programming once I feel like more of an infant in the subject.
The next programming blog post will likely be the affirmation and fun-fact smartwatch automation I have been using for the last few months, or a Kismet dashboard with all the metadata deployed in Kibana. That may be a future project because my Kali hardware instance was hosed/foobared/corrupted soon after the completion of another related project.
Now back to the subject at hand: our intel agent. This agent has a few different components. There is an agent that will scrape documents from the website if its knowledge stores are not sufficient enough, or if you ask it to scrape the website. For instance, say you would like to update the local cache to include news from the last month—you could run a query like <scrape news data for the last month>. This can be done on the web app or on the CLI.
Right now, the data gathered or available scrapers are Countries, Vehicles, Weapons, Hot Documents, News, Reports, WMD Countries, Operations, Intelligence Systems, or you can run all the scrapers at once. Once loaded locally, you have the ability to load this information into a vector database. The local store is best for exact keyword matches like country names, versus a vector DB being better with context-based searching. The vector DB is better at handling synonyms and understanding meaning and context.
The agent will even take your query and enhance it by forming a new research query and then answering the new query, or it will answer the raw question. The research query is not deterministic and therefore creates some problems where sometimes the result is exceptional and other times it broadens the query and instead of just giving the treaties between the USA and the UAE, it also includes those of the USA and Japan. This just needs refinement in another iteration.
Disclaimer this agent is not better than reading the information for the site itself. So for instance the idea of simulating a conflict between Greenland and the USA has already been done in far more detail by experts in the field on Global Security. So where I would use this tool is as really efficient news search tool to find news stories similar to the subject I was interested in that aren’t gathered by the advance search or to quick analyst facts like who has the most advanced weapons systems or spends more on the defense budget.
Leave a Reply