Introduction

​

Welcome

Understand how to use ScholarAI, where it is limited, and where we are developing.

​

Suggested Starting Steps

​

Known Limitations

As we have been developing for tooling that is being used by LLMs and agents, there a number of known spaces for improvement as it relates to developer usage and expecations of an API

• Most critically, the speed of search is not fast for ad-hoc queries and UI rendering - our search endpoint will scan our vector database via hybrid search, a backfill for real-time open access data, and optionally pass in the discovered data into a LLM for synthesizing answers. In the context of our GPT and LLM tools, users have come to expect latency as a tradeoff for feature extensibility, but as a developer tool this is a prominent space for improvement.
• Many endpoints have default behavior or overrides not captured in the specification for the sake of brevity for models with small context windows. We will continue to remove these hidden patterns and explore implementations that make endpoints easy to use both by LLMs and developers.

We are open to all feedback as we work on making this system better every day!