I built a comprehensive user interface and front end that works with several of the major large language models (LLMs), including OpenAI's ChatGPT, Anthropic's Claude, X's Grok and Google's Gemini. My goal was specifically to build a front-end to facilitate scientific use. This entailed building a front end that lends itself to good quality PDF generation.

More importantly, the front-end is integrated with the Maxima computer algebra system, which permits the LLM to carry out complex calculations and plot functions. The LLM also has access to Google.

Unlike more recent "function call" implementations, WISPL grants agency to the model itself to decide if/when it responds with a query instead of directly answering the user's prompt. This leads to a more natural conversation flow.

More recently, I added the capability to process images, and also the capability to process PDF files as images. This is particularly useful in the scientific-technical context: PDF file contents are badly mangled by text conversion, whereas all modern frontier-class models are multimodal, can process image content, and are perfectly capable of reading even heavily distorted text quite accurately.

I may never commercialize this. The business model is shaky, there are security implications, and the market is crowded. Nonetheless, I find it important to showcase, as it is a quality solution that, at least in certain contexts, may serve its users better than the "official" interfaces offered by the respective AI service providers.