Research

At Kapa, we invest deeply in understanding the technical foundations that power our platform. This section shares findings from our internal research, including benchmarks, evaluations, and engineering insights that shape how we build Kapa.

These posts are not product guides. They are a window into the rigor behind our engineering decisions, published so our customers and the broader community can benefit from what we learn.

How to Properly Evaluate a RAG-Based AI Agent· 12mFinn Bauer · May 11, 2026Most vendor evaluations produce meaningless results. We share the methodology we use internally — structured test datasets scored by LLM-as-a-judge — so you can run an evaluation that actually tells you something.

Why We Built Our Own PDF Converter Benchmark· 15mLars Baltensperger · April 13, 2026We evaluated leading PDF converters across heading hierarchy, table extraction, figure detection, and text fidelity to find the best fit for our ingestion pipeline.