James Gao

Liya Wang, David Yi, Damien Jose et al.

Large Language Models (LLMs) ) have demonstrated promise in boosting productivity across AI-powered tools, yet existing benchmarks like Massive Multitask Language Understanding (MMLU) inadequately assess enterprise-specific task complexities. We propose a 14-task framework grounded in Bloom's Taxonomy to holistically evaluate LLM capabilities in enterprise contexts. To address challenges of noisy data and costly annotation, we develop a scalable pipeline combining LLM-as-a-Labeler, LLM-as-a-Judge, and corrective retrieval-augmented generation (CRAG), curating a robust 9,700-sample benchmark. Evaluation of six leading models shows open-source contenders like DeepSeek R1 rival proprietary models in reasoning tasks but lag in judgment-based scenarios, likely due to overthinking. Our benchmark reveals critical enterprise performance gaps and offers actionable insights for model optimization. This work provides enterprises a blueprint for tailored evaluations and advances practical LLM deployment.

Dylan Lester, James Gao, Samuel Sutphin et al.

Manual labeling for large-scale image and video datasets is often time-intensive, error-prone, and costly, posing a significant barrier to efficient machine learning workflows in fault detection from railroad videos. This study introduces a semi-automated labeling method that utilizes a pre-trained You Only Look Once (YOLO) model to streamline the labeling process and enhance fault detection accuracy in railroad videos. By initiating the process with a small set of manually labeled data, our approach iteratively trains the YOLO model, using each cycle's output to improve model accuracy and progressively reduce the need for human intervention. To facilitate easy correction of model predictions, we developed a system to export YOLO's detection data as an editable text file, enabling rapid adjustments when detections require refinement. This approach decreases labeling time from an average of 2 to 4 minutes per image to 30 seconds to 2 minutes, effectively minimizing labor costs and labeling errors. Unlike costly AI based labeling solutions on paid platforms, our method provides a cost-effective alternative for researchers and practitioners handling large datasets in fault detection and other detection based machine learning applications.

James Gao, Josh Zhou, Qi Sun et al.

Many tool-based Retrieval Augmented Generation (RAG) systems lack precise mechanisms for tracing final responses back to specific tool components -- a critical gap as systems scale to complex multi-agent architectures. We present \textbf{Atomic Information Flow (AIF)}, a graph-based network flow model that decomposes tool outputs and LLM calls into atoms: indivisible, self-contained units of information. By modeling LLM orchestration as a directed flow of atoms from tool and LLM nodes to a response super-sink, AIF enables granular attribution metrics for AI explainability. Motivated by the max-flow min-cut theorem in network flow theory, we train a lightweight Gemma3 (4B parameter) language model as a context compressor to approximate the minimum cut of tool atoms using flow signals computed offline by AIF. We note that the base Gemma3-4B model struggles to identify critical information with \textbf{54.7\%} accuracy on HotpotQA, barely outperforming lexical baselines (BM25). However, post-training on AIF signals boosts accuracy to \textbf{82.71\%} (+28.01 points) while achieving \textbf{87.52\%} (+1.85\%) context token compression -- bridging the gap with the Gemma3-27B variant, a model nearly $7\times$ larger.