Ambient Field Lens

Ambient Field Lens investigates whether an AI can recognise gaps in its knowledge—and turns those gaps into documentation a human can review, improve and deploy. During commissioning and troubleshooting, the crucial instruction is often missing or scattered across manuals and experienced colleagues. A fluent AI answer can hide that gap, leading to unnecessary repairs, longer downtime and repeat visits. Our prototype combines a camera-equipped iOS app, document retrieval, visual segmentation and an evidence-bound response. We also read Glimmer’s internal activations before it generates an answer to explore whether they reveal missing support in the retrieved documents. The system flags possible documentation gaps through retrieval checks and keeps the technician in control: every response requires verification. The demonstration uses a drawing robot. An operator asks what to do when the paper is incorrectly positioned, examines the retrieved evidence, and writes the missing procedure in a wiki. After human approval and re-indexing, the identical question cites the new passage. Both corpus versions remain reproducible, with distinct citation IDs and a rollback path. The result is a documentation improvement loop: inspect the evidence, supply the missing knowledge, and make it available to the next technician—without retraining the model. Demo and technical evidence: https://afl.flotilla.cc/