arXiv subjects

Keertan Balaji

Publications and source records attributed to Keertan Balaji.

4 recordsLinked to original sources

arXiv2026

OpenDiscoveryTrace: Process Traces for Evaluating AI Scientist Workflows

Existing benchmarks for autonomous AI scientists evaluate only final outputs---generated code, hypotheses, or papers---yet discard the reasoning process by which those outputs were obtained. This makes it impossible to audit scientific methodology, diagnose failure modes, or distinguish systematic reasoning from fortunate guessing. We present \textbf{OpenDiscoveryTrace}, a public dataset of 558 complete AI scientific agent trajectories that captures how models reason, not just what they produce. Each trajectory records a structured 9-field-per-step trace---including thoughts, tool calls, observations, errors, revision triggers, and self-reported confidence---as models execute 124 scientific tasks spanning drug discovery, materials science, genomics, and scientific literature analysis. The dataset covers seven models: three frontier models (GPT-5.4, Claude Opus 4.6, and Gemini 3.1 Pro; 124 trajectories each, fully balanced across domains and difficulty levels) and four open-weight models (Qwen2.5-7B, Mistral-7B-v0.3, Phi-3.5-mini, and Qwen2.5-1.5B; 30 each), plus 60 live-retrieval variant trajectories. Pilot analysis on 363 LLM-judged trajectories reveals that process traces expose behavioral differences invisible to output-only evaluation: all three frontier models achieve comparable success rates (84--89%), yet Claude Opus 4.6 produces 30$\times$ more errors than GPT-5.4 (2.5 vs. 0.08 per trajectory, $p < 0.0001$, Cliff's $δ= 0.613$), with qualitatively different error profiles---66.7% tool misuse for Claude versus 83.6% reasoning errors for GPT-5.4. We define five benchmark tasks with baselines from logistic regression, random forests, LSTMs, and Transformer models. The dataset, trace schema, agent harness, and benchmark definitions are publicly available under CC BY 4.0 to support research on process-level evaluation, scientific agent auditing, and AI governance.

cs.AI↗

arXiv2025

A Recall-First CNN for Sleep Apnea Screening from Snoring Audio

Sleep apnea is a serious sleep-related breathing disorder that is common and can impact health if left untreated. Currently the traditional method for screening and diagnosis is overnight polysomnography. Polysomnography is expensive and takes a lot of time, and is not practical for screening large groups of people. In this paper, we explored a more accessible option, using respiratory audio recordings to spot signs of apnea.We utilized 18 audio files.The approach involved converting breathing sounds into spectrograms, balancing the dataset by oversampling apnea segments, and applying class weights to reduce bias toward the majority class. The model reached a recall of 90.55 for apnea detection. Intentionally, prioritizing catching apnea events over general accuracy. Despite low precision,the high recall suggests potential as a low-cost screening tool that could be used at home or in basic clinical setups, potentially helping identify at-risk individuals much earlier.

cs.SD↗

arXiv2025

Temporal Encoding Strategies for Energy Time Series Prediction

In contemporary power systems, energy consumption prediction plays a crucial role in maintaining grid stability and resource allocation enabling power companies to minimize energy waste and avoid overloading the grid. While there are several research works on energy optimization, they often fail to address the complexities of real-time fluctuations and the cyclic pattern of energy consumption. This work proposes a novel approach to enhance the accuracy of predictive models by employing sinusoidal encoding on periodic features of time-series data. To demonstrate the increase in performance, several statistical and ensemble machine learning models were trained on an energy demand dataset, using the proposed sinusoidal encoding. The performance of these models was then benchmarked against identical models trained on traditional encoding methods. The results demonstrated a 12.6% improvement of Root Mean Squared Error (from 0.5497 to 0.4802) and a 7.8% increase in the R^2 score (from 0.7530 to 0.8118), indicating that the proposed encoding better captures the cyclic nature of temporal patterns than traditional methods. The proposed methodology significantly improves prediction accuracy while maintaining computational efficiency, making it suitable for real-time applications in smart grid systems.

cs.LG↗

arXiv2025

VocalEyes: Enhancing Environmental Perception for the Visually Impaired through Vision-Language Models and Distance-Aware Object Detection

With an increasing demand for assistive technologies that promote the independence and mobility of visually impaired people, this study suggests an innovative real-time system that gives audio descriptions of a user's surroundings to improve situational awareness. The system acquires live video input and processes it with a quantized and fine-tuned Florence-2 big model, adjusted to 4-bit accuracy for efficient operation on low-power edge devices such as the NVIDIA Jetson Orin Nano. By transforming the video signal into frames with a 5-frame latency, the model provides rapid and contextually pertinent descriptions of objects, pedestrians, and barriers, together with their estimated distances. The system employs Parler TTS Mini, a lightweight and adaptable Text-to-Speech (TTS) solution, for efficient audio feedback. It accommodates 34 distinct speaker types and enables customization of speech tone, pace, and style to suit user requirements. This study examines the quantization and fine-tuning techniques utilized to modify the Florence-2 model for this application, illustrating how the integration of a compact model architecture with a versatile TTS component improves real-time performance and user experience. The proposed system is assessed based on its accuracy, efficiency, and usefulness, providing a viable option to aid vision-impaired users in navigating their surroundings securely and successfully.

cs.HC↗