Search papers, labs, and topics across Lattice.
The paper introduces Loom, a generative consensus framework that effectively aggregates conflicting textual hypotheses for Root Cause Analysis (RCA) by utilizing an iterative centroid-based reweighting algorithm in a continuous embedding space. This approach allows Loom to achieve a balance between accuracy and efficiency, outperforming traditional methods by being approximately 26 times faster while maintaining competitive accuracy against state-of-the-art autonomous agents. The findings highlight the practical implications of Loom in real-world industrial settings, demonstrating its ability to foster trust among Subject Matter Experts through deterministic consensus.
Loom achieves a remarkable 26x speedup in Root Cause Analysis while maintaining competitive accuracy, redefining efficiency in NLP applications.
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).
NVIDIA