Tanvir Bhathal

Jon Saad-Falcon, Avanika Narayan, Robby Manihani et al.

Personal AI stacks, like OpenClaw and Hermes Agent, are becoming central to daily work, yet they route nearly every query (often over sensitive local data) to cloud-hosted frontier models. Replacing frontier models with local models inside existing stacks does not work: swapping Claude Opus 4.6 for Qwen3.5-9B drops accuracy by 25-39 pp across personal AI tasks like PinchBench and GAIA. Existing stacks bundle agentic prompts, tool descriptions, memory configuration, and runtime settings around a specific cloud model. Only the prompts can be tuned, and state-of-the-art prompt optimizers close just 5 pp of the local-cloud gap on their own. This motivates a decomposed personal AI stack: one that exposes individual primitives which can be optimized individually or jointly to close the local-cloud gap. We present OpenJarvis, an architecture that represents a personal AI system as a typed spec over five primitives: Intelligence, Engine, Agents, Tools & Memory, and Learning. Each primitive is an independently editable field, making the stack end-to-end optimizable and measurable against accuracy, cost, and latency. Towards closing the local-cloud gap without surrendering local-model properties, OpenJarvis introduces LLM-guided spec search, a local-cloud collaboration in which frontier cloud models propose edits across the spec at search time, only non-regressing edits are accepted, and the resulting spec runs entirely on-device at inference time. With LLM-guided spec search, on-device specs match or exceed cloud accuracy on 4 of 8 benchmarks and land within 3.2 pp of the best cloud baseline on average. They also reduce marginal API cost by ~800x and end-to-end latency by 4x.

Prady Saligram, Tanvir Bhathal, Robby Manihani

Delayed and sparse rewards present a fundamental obstacle for reinforcement-learning (RL) agents, which struggle to assign credit for actions whose benefits emerge many steps later. The sliding-tile game 2048 epitomizes this challenge: although frequent small score changes yield immediate feedback, they often mislead agents into locally optimal but globally suboptimal strategies. In this work, we introduce a unified, distributional multi-step RL framework designed to directly optimize long-horizon performance. Using the open source Gym-2048 environment we develop and compare four agent variants: standard DQN, PPO, QR-DQN (Quantile Regression DQN), and a novel Horizon-DQN (H-DQN) that integrates distributional learning, dueling architectures, noisy networks, prioritized replay, and more. Empirical evaluation reveals a clear hierarchy in effectiveness: max episode scores improve from 3.988K (DQN) to 5.756K (PPO), 8.66K (QR-DQN), and 18.21K (H-DQN), with H-DQN reaching the 2048 tile. Upon scaling H-DQN it reaches a max score 41.828K and a 4096 tile. These results demonstrate that distributional, multi-step targets substantially enhance performance in sparse-reward domains, and they suggest promising avenues for further gains through model-based planning and curriculum learning.

Jon Saad-Falcon, E. Kelly Buchanan, Mayee F. Chen et al.

Verifiers can improve language model capabilities by scoring and ranking responses from generated candidates. Currently, high-quality verifiers are either unscalable (e.g., humans) or limited in utility (e.g., tools like Lean). While LM judges and reward models have become broadly useful as general-purpose verifiers, a significant performance gap remains between them and oracle verifiers (verifiers with perfect accuracy). To help close this gap, we introduce Weaver, a framework for designing a strong verifier by combining multiple weak, imperfect verifiers. We find weighted ensembles of verifiers, which typically require learning from labeled data, significantly outperform unweighted combinations due to differences in verifier accuracies. To reduce dependency on labeled data, Weaver leverages weak supervision to estimate each verifier's accuracy and combines outputs into a unified score that better reflects true response quality. However, directly applying weak supervision algorithms poses challenges, including inconsistent verifier output formats and handling low-quality verifiers. Weaver addresses these using dataset statistics to normalize outputs and filter specific verifiers. We study Weaver's effectiveness in test-time repeated sampling, where a model generates multiple candidate responses and selects one. Our evaluations show Weaver significantly improves over Pass@1-performance when selecting the first candidate-across reasoning and math tasks, achieving o3-mini-level accuracy with Llama 3.3 70B Instruct as generator, and an ensemble of 70B or smaller judge and reward models as verifiers (87.7% average). This gain mirrors the jump between GPT-4o and o3-mini (69.0% vs. 86.7%), which required extensive finetuning and post-training. To reduce computational costs of verifier ensembles, we train a 400M cross-encoder using Weaver's combined output scores.

Tanvir Bhathal, Asanshay Gupta

We introduce WebSight, a vision-based autonomous web agent, designed to interact with web environments purely through visual perception, eliminating dependence on HTML or DOM-based inputs. Central to our approach we introduce our new model, WebSight-7B, a fine-tuned vision-language model optimized for UI element interaction, trained using LoRA on a web-focused subset of the Wave-UI-25K dataset. WebSight integrates this model into a modular multi-agent architecture, comprising planning, reasoning, vision-action, and verification agents, coordinated through an episodic memory mechanism. WebSight-7B achieves a top-1 accuracy of 58.84% on the Showdown Clicks benchmark, outperforming several larger generalist models while maintaining lower latency. The full WebSight agent achieves a 68.0% success rate on the WebVoyager benchmark, surpassing systems from labs such as OpenAI (61.0%) and HCompany (Runner H, 67.0%). Among tasks completed, WebSight answers correctly 97.14% of the time, indicating high precision. Together, WebSight and WebSight-7B establish a new standard for interpretable, robust, and efficient visual web navigation.

Prady Saligram, Tanvir Bhathal

Air pollution from agricultural emissions is a significant yet often overlooked contributor to environmental and public health challenges. Traditional air quality forecasting models rely on physics-based approaches, which struggle to capture complex, nonlinear pollutant interactions. In this work, we explore forecasting N$_2$O agricultural emissions through evaluating popular architectures, and proposing two novel deep learning architectures, EmissionNet (ENV) and EmissionNet-Transformer (ENT). These models leverage convolutional and transformer-based architectures to extract spatial-temporal dependencies from high-resolution emissions data