The technical foundation of Bigspin is the open-source DSPy project, which is led by Chris’s former PhD student Omar Khattab and which grew out of their research collaborations with Matei Zaharia, Heather Miller, and many other scholars and open-source developers. 

In DSPy, prompt engineering is replaced by foundation model programming. At the lowest level, this means that systems are expressed in modular computer code rather than with fragile hand-crafted prompt templates. To the extent that prompts are written by hand at all, it is just to express the central goals and requirements of the system. The details of formatting, output typing, and message passing are left to DSPy, in much the same way that traditional software compilers handle the translation from high-level programming languages to machine code.

At Bigspin, DSPy is also our primary tool for data-driven optimization. All too often, GenAI developers end up iterating on prompts based on vibes and a few examples. Such iteration is exceedingly unlikely to lead to an optimal system; no one would try to set the weights of a neural network by hand, but even experienced AI developers end up doing something akin to this with their prompts.

One of DSPy’s key innovations was to replace these inefficient and uncertain development patterns with proper data-driven optimization: given a set of labeled examples and an initial system specification, DSPy’s optimizers seek to find a prompt that maximizes performance on those examples. In its most general form, this is a highly intractable optimization problem, but the research community has discovered approximations of it that are effective in practice, and DSPy capitalizes on these discoveries. In head-to-head comparisons, DSPy optimizers beat hand-crafted prompts every time.

Optimizers are not magical, though, and they are certainly not substitutes for domain expertise. At Bigspin, one of our guiding principles is that GenAI development also depends on a discovery process in which the subject matter experts and other stakeholders collectively figure out what good looks like for their application: how it should feel, what its requirements are, and so forth. Bigspin facilitates this discovery process and synthesizes all of the resulting labeled examples and other feedback. DSPy optimizers then use all that information to create a tailored and trusted prompt that is transformatively better than any prompt one could find by hand.

Bigspin’s applications provide all the benefits of using DSPy, but with no technical overhead. Our central mission is to enable individuals and teams to concentrate on applying their domain expertise to the task of developing the best possible GenAI solutions.