Original title: Loom: Weaving Diagnostic Strands into Free-Text Consensus via Embedding-Space Reweighting
Ron Begleiter, Katya Egert Berg, Gilad Saban, Gil Shabat · September 2026
When complex industrial systems break down, finding the root cause is difficult. Big artificial intelligence models can help, but they make mistakes, hit memory limits, and take too long to run. Meanwhile, simpler automated rules generate noisy and conflicting guesses. Researchers created Loom to solve this. Loom collects free-text guesses from diagnostic templates filled with specific incident details like times, entities, and metrics. It turns these guesses into mathematical points in a shared space and uses an algorithm to reweight them until it finds the central, most reliable points. Finally, a single lightweight language model writes a final summary from these weighted ideas. In real tests, Loom ran approximately 26 times faster than a top autonomous agent, matched its accuracy on two benchmarks, and trailed on two others.
Imagine you're packing for a trip, and several friends give different advice on what to bring based on the weather forecast. It's hard to know who to listen to because their suggestions conflict. In this paper, Loom is like a method that groups similar packing tips, finds the most common items, and uses a quick decision to finalize the list without lengthy arguments.
In industrial settings, teams waste time reconciling different expert opinions during system failures. Loom automates this, giving a reliable consensus much faster, which speeds up fixes and reduces downtime.
Loom does not outperform state-of-the-art agents on all datasets, as it trails on Market-1 and Telecom benchmarks. It had negative results in detecting redundant information. The work is focused on root cause analysis in NLP systems, not broader applications.
Ask anything about this paper. Answers come from its abstract, and say so when the abstract doesn't cover your question.
Loom: Weaving Diagnostic Strands into Free-Text Consensus via Embedding-Space Reweighting — Ron Begleiter, Katya Egert Berg, Gilad Saban et al.
Abstract licence: CC0
Explanation generated by Synthesis from the abstract — it is not the original paper. This is a preprint, so it hasn't necessarily been peer reviewed.