Loom: Weaving Diagnostic Strands into Free-Text Consensus via Embedding-Space Reweighting
Ron Begleiter, Katya Egert Berg, Gilad Saban and Gil Shabat at NVIDIA present Loom, a deployed root cause analysis system that aggregates open-form hypotheses from modular heuristics in embedding space and spends exactly one LLM call per incident.
Ask this paper
The tradeoff being made explicit: A monolithic LLM agent has unbounded expressivity and pays for it in context limits, compounding hallucinations and inference latency. Classical weak supervision has statistical rigor but is mathematically restricted to discrete classes, so it cannot aggregate free text at all.
Consensus in embedding space: Diagnostic templates are populated with episode-specific entities, times and metrics to emit open-form hypotheses. Loom projects them into a continuous embedding space and resolves conflicts with an iterative centroid-based reweighting algorithm, then uses the resulting weights to ground a single lightweight LLM synthesis step.
Pareto frontier on OpenRCA: Loom matches a state-of-the-art autonomous agent on Bank and Market-2 and trails on Market-1 and Telecom, using one LLM call per incident on all four datasets, roughly 26x faster and about 33x with an 8B synthesizer.
Deployment lessons, including negative ones: The paper reports the tradeoff between agentic depth and inference latency, negative results in redundancy detection, and the observation that deterministic consensus builds more trust with subject matter experts than an agent whose output varies run to run.
Why it matters: A counterweight to the assumption that more agentic depth is always better. On a real industrial diagnosis workload, most of the accuracy survives collapsing the agent loop to a single synthesis call, and the determinism is what makes domain experts willing to use it.
Abstract
Aggregating noisy, conflicting textual hypotheses into a reliable consensus is a fundamental challenge when deploying NLP systems in real-world industrial settings. While monolithic Large Language Model (LLM) agents offer unbounded expressivity for tasks like Root Cause Analysis (RCA), they suffer from context limits, compounding hallucinations, and prohibitive inference latency. Traditional weak supervision offers statistical rigor but is mathematically restricted to discrete classes. We present Loom, a generative consensus framework deployed for real-world RCA that bridges these paradigms. Loom aggregates open-form hypotheses emitted by modular heuristics (diagnostic templates dynamically populated with episode-specific entities, times, and metrics) by projecting them into a continuous embedding space, and resolves conflicting signals with an iterative centroid-based reweighting algorithm. The resulting consensus weights ground a single lightweight LLM synthesis step. Evaluated on the OpenRCA benchmark, Loom occupies the accuracy--efficiency Pareto frontier: it matches a state-of-the-art autonomous agent on Bank and Market-2 and trails on Market-1 and Telecom, while using a single LLM call per incident on all four datasets ($\sim$26$\times$ faster; $\sim$33$\times$ with an 8B-parameter synthesizer). We discuss our deployment experience, highlighting lessons learned regarding the trade-offs between agentic depth and inference latency, negative results in redundancy detection, and how deterministic consensus fosters trust among Subject Matter Experts~(SMEs).