Dissertation·In progress
Evaluating Retrieval-Augmented Systems
Research into how retrieval systems surface evidence, preserve context and fail under realistic organisational conditions.
RAG · Evaluation · Institutional knowledge
Research
Questions that
refuse to stay
theoretical.
Research on retrieval, evidence, institutional memory and the conditions under which intelligent systems become trustworthy enough to matter.
Some work here is formal academic research. Some is exploratory. The distinction will always be made clear.
Current research
Work in progress, active inquiry and research notes that may later become papers, essays or larger projects.
Research note·Developing
What Retrieval Systems Reveal About Institutional Memory
A working inquiry into the difference between information being stored and knowledge remaining meaningfully available.
Retrieval · Memory · Knowledge systems
Method·Open question
Evidence Before Fluency
How should we evaluate a system when a plausible answer may still be built on weak, partial or misleading evidence?
Evaluation · Evidence · AI systems
Questions I am carrying
Not every useful question needs an immediate answer. Some are more valuable when held open long enough to change the work.
01
When does retrieval become interpretation?
02
What should count as evidence when an answer is assembled from many sources?
03
How much institutional context is lost before information becomes technically retrievable?
04
Can a system be useful while still being epistemically unreliable?
Method
Evidence before confidence.
I am interested in systems that can explain not only what they answer, but what evidence made the answer possible.
That means treating retrieval quality, provenance, uncertainty, failure modes and institutional context as first-class parts of evaluation rather than secondary implementation details.
Read related essay →