After just over four months, I now have a solid, well-developed core of knowledge-based, tech-powered solutions supporting my two main research areas, both related to Cuba–United States relations: 1) Washington's persistent efforts to overthrow Cuba's political system; and 2) the architecture of the sanctions regime imposed on the Island since 1960.
While preparing for my final doctoral dissertation defense last June, I began developing and documenting on Hive a set of tools built with Python and JavaScript. The first was a simple search engine designed merely to replicate the searches I perform on USASpending.gov, a federal spending database that tracks the allocation and obligation of government funds.
A great deal has happened since June 22, when I first wrote about that initial take that has now become part of the Cuba Democracy Hub. So far, the Hub is powered by a grants-tracking program and provides unprecedented traceability across previously independent datasets.
Thus, it includes a more sophisticated search of USASpending records than I originally envisioned, as well as the records available through FAC.gov. The latter involves both the platform's API—which contains the SF-SAC forms submitted by organizations that expend more than $750,000 annually in federal assistance funds—and the PDF audit reports that provide detailed breakdowns of the data contained in those forms.
This last component is arguably the most significant achievement so far. It parses a collection of non-standardized documents, a particularly demanding technical challenge. The next step will be to integrate several additional components into the Hub:
The pipeline described in A Modular Python Pipeline to..., which monitors the apportionments released by the Office of Management and Budget to the agencies administering the Cuba program.
The grants search engine, which will become a more visible and accessible component of the platform.
The ForeignAssistance.gov tracker, which should, in principle, match its records against USASpending data. In any case, the two sources are complementary and provide additional analytical value.
If I manage to bring all this to fruition, it could open the door to leading a major research project, given its strategic importance to national security. Throughout this effort, I have relied consistently on LLMs such as Claude, DeepSeek, ChatGPT, and Grok, all in their chatbot interfaces.
It hasn't been an easy journey, for two main reasons. First, working with these systems is not as simple as it might seem. It requires engineering knowledge and skills, along with a strong commitment to quality assurance, to subject both code and functionality to rigorous testing.
Second, there is a chronic problem facing Cubans, rooted in our persistent economic hardship and the discrimination we experience when trying to access these AI services, let alone pay for subscription plans that would allow us to make more substantial use of them.
Code generation is particularly demanding in terms of token consumption, and those tokens disappear mercilessly fast. This is especially true with Claude, which, among all the systems I have worked with—including Gemini, Copilot, and Kimi—has proven to be the most comprehensive and effective, as well as the most straightforward to work with.
I'll leave you with two more snapshots of my Cuba Democracy Hub until my next contribution, which will be dedicated to my Cuba Sanctions Hub.