arXiv subjects

Matthew Russo

Publications and source records attributed to Matthew Russo.

17 recordsLinked to original sources

arXiv2026

Carnot: Interpretable, Interactive, and Optimized Execution of Deep Research Queries

Enterprises increasingly seek to query data lakes using natural language via AI-driven tools like semantic operators or deep research agents. However, the latter operates as an opaque black box, hiding its intermediate reasoning and data retrieval steps, and failing to expose controls for managing API costs and execution latency. Meanwhile, the former can be prohibitively expensive for enterprise-scale data lakes. Consequently, analysts using these systems lack the agency to intercept hallucinated premises, verify intermediate results, or correct the system's trajectory. We present Carnot, an interactive execution engine for AI-driven analytics. Carnot compiles natural language requests into physical execution graphs and surfaces them through an interactive notebook interface. Rather than waiting blindly for a final output, users can critique the plan, incrementally execute operators, inspect intermediate data, or directly edit the underlying code or semantic operator instructions. Carnot's query optimizer will optimize the query with respect to cost or latency constraints provided by the user. Our demo will showcase how Carnot helps users achieve efficient and verifiable insights on workloads motivated by real enterprise use cases.

cs.DB↗

arXiv2026

SAGE: Selective Attention-Guided Extraction for Token-Efficient Document Indexing

Large language models with long context windows can answer complex questions directly from full-length academic, technical, and policy documents, but passing entire documents is often costly, slow, and can degrade answer quality while increasing the risk of unnecessary data leakage. This paper targets the common setting of answering many heterogeneous questions over long document(s), where fixed position heuristics and standard retrieval-augmented generation (RAG) can fail due to document structure variability and weak query-chunk semantic similarity, which often requires task- and domain-specific tuning of embedding retrievers. We propose {Selective Attention-Guided Extraction} (\ourmethod), a training-free, plug-and-play context reduction framework that uses a lightweight local LLM to perform a single prefilling pass and convert language model attention signals into a query-specific relevance heatmap at configurable granularities. \ourmethod\ further introduces \emph{differential attention} strategies to better isolate question-relevant evidence, then selects the top-scoring units under a user-defined token budget and forwards only this reduced context to a downstream LLM for answer generation. \ourmethod\ surpasses traditional reduction techniques across multiple long-document QA benchmarks, notably securing a top-4 rank on QuALITY-hard while constrained to a 10\% context budget. This enables a 90\% reduction in tokens with competitive accuracy, without the need for model fine-tuning or complex calibration.

cs.DB↗

arXiv2026

Agent-Aided Design for Dynamic CAD Models

In the past year, researchers have created agentic systems that can design real-world CAD-style objects in a training-free setting, a new variety of system that we call Agent-Aided Design. These systems place an agent in a feedback loop in which it generates an assembly of CAD model(s), visualizes the assembly, and then iteratively refines its assembly based on visual and other feedback. Despite rapid progress, a key problem remains: none of these systems can build complex 3D assemblies with moving parts. For example, no existing system can build a piston, a pendulum, or even a pair of scissors. In order for Agent-Aided Design to make a real impact in industrial manufacturing, we need a system that is capable of generating such 3D assemblies. In this paper we present a prototype of AADvark, an agentic system designed for this task. Unlike previous state-of-the-art systems, AADvark captures the dynamic part interactions with one or more degrees-of-freedom. This design decision allows AADvark to reason directly about assemblies with moving parts and can thereby achieve cross-cutting goals, including but not limited to mechanical movements. Unfortunately, current LLMs are imperfect spatial reasoners, a problem that AADvark addresses by incorporating external constraint solver tools with a specialized visual feedback mechanism. We demonstrate that, by modifying the agent's tools (FreeCAD and the assembly solver), we are able to create a strong verification signal which enables our system to build 3D assemblies with movable parts.

cs.AI↗

arXiv2025

SemBench: A Benchmark for Semantic Query Processing Engines

We present a benchmark targeting a novel class of systems: semantic query processing engines. Those systems rely inherently on generative and reasoning capabilities of state-of-the-art large language models (LLMs). They extend SQL with semantic operators, configured by natural language instructions, that are evaluated via LLMs and enable users to perform various operations on multimodal data. Our benchmark introduces diversity across three key dimensions: scenarios, modalities, and operators. Included are scenarios ranging from movie review analysis to car damage detection. Within these scenarios, we cover different data modalities, including images, audio, and text. Finally, the queries involve a diverse set of operators, including semantic filters, joins, mappings, ranking, and classification operators. We evaluated our benchmark on three academic systems (LOTUS, Palimpzest, and ThalamusDB) and one industrial system, Google BigQuery. Although these results reflect a snapshot of systems under continuous development, our study offers crucial insights into their current strengths and weaknesses, illuminating promising directions for future research.

cs.DB↗

arXiv2025

Deep Research is the New Analytics System: Towards Building the Runtime for AI-Driven Analytics

With advances in large language models (LLMs), researchers are creating new systems that can perform AI-driven analytics over large unstructured datasets. Recent work has explored executing such analytics queries using semantic operators -- a declarative set of AI-powered data transformations with natural language specifications. However, even when optimized, these operators can be expensive to execute on millions of records and their iterator execution semantics make them ill-suited for interactive data analytics tasks. In another line of work, Deep Research systems have demonstrated an ability to answer natural language question(s) over large datasets. These systems use one or more LLM agent(s) to plan their execution, process the dataset(s), and iteratively refine their answer. However, these systems do not explicitly optimize their query plans which can lead to poor plan execution. In order for AI-driven analytics to excel, we need a runtime which combines the optimized execution of semantic operators with the flexibility and more dynamic execution of Deep Research systems. As a first step towards this vision, we build a prototype which enables Deep Research agents to write and execute optimized semantic operator programs. We evaluate our prototype and demonstrate that it can outperform a handcrafted semantic operator program and open Deep Research systems on two basic queries. Compared to a standard open Deep Research agent, our prototype achieves up to 1.95x better F1-score. Furthermore, even if we give the agent access to semantic operators as tools, our prototype still achieves cost and runtime savings of up to 76.8% and 72.7% thanks to its optimized execution.

cs.AI↗

arXiv2025

KramaBench: A Benchmark for AI Systems on Data-to-Insight Pipelines over Data Lakes

Discovering insights from a real-world data lake potentially containing unclean, semi-structured, and unstructured data requires a variety of data processing tasks, ranging from extraction and cleaning to integration, analysis, and modeling. This process often also demands domain knowledge and project-specific insight. While AI models have shown remarkable results in reasoning and code generation, their abilities to design and execute complex pipelines that solve these data-lake-to-insight challenges remain unclear. We introduce KramaBench which consists of 104 manually curated and solved challenges spanning 1700 files, 24 data sources, and 6 domains. KramaBench focuses on testing the end-to-end capabilities of AI systems to solve challenges which require automated orchestration of different data tasks. KramaBench also features a comprehensive evaluation framework assessing the pipeline design and individual data task implementation abilities of AI systems. We evaluate 8 LLMs using our single-agent reference framework DS-Guru, alongside both open- and closed-source single- and multi-agent systems, and find that while current agentic systems may handle isolated data-science tasks and generate plausible draft pipelines, they struggle with producing working end-to-end pipelines. On KramaBench, the best system reaches only 55% end-to-end accuracy in the full data-lake setting. Even with perfect retrieval, the accuracy tops out at 62%. Leading LLMs can identify up to 42% of important data tasks but can only fully implement 20% of individual data tasks. Our code, reference framework, and data are available at https://github.com/mitdbg/KramaBench.

cs.DB↗

arXiv2025

Abacus: A Cost-Based Optimizer for Semantic Operator Systems

LLMs enable an exciting new class of data processing applications over large collections of unstructured documents. Several new programming frameworks have enabled developers to build these applications by composing them out of semantic operators: a declarative set of AI-powered data transformations with natural language specifications. These include LLM-powered maps, filters, joins, etc. used for document processing tasks such as information extraction, summarization, and more. While systems of semantic operators have achieved strong performance on benchmarks, they can be difficult to optimize. An optimizer for this setting must determine how to physically implement each semantic operator in a way that optimizes the system globally. Existing optimizers are limited in the number of optimizations they can apply, and most (if not all) cannot optimize system quality, cost, or latency subject to constraint(s) on the other dimensions. In this paper we present Abacus, an extensible, cost-based optimizer which searches for the best implementation of a semantic operator system given a (possibly constrained) optimization objective. Abacus estimates operator performance by leveraging a minimal set of validation examples, prior beliefs about operator performance, and/or an LLM judge. We evaluate Abacus on document processing workloads in the biomedical and legal domains (BioDEX; CUAD) and multi-modal question answering (MMQA). We demonstrate that, on-average, systems optimized by Abacus achieve 6.7%-39.4% better quality and are 10.8x cheaper and 3.4x faster than the next best system.

cs.DB↗

arXiv2024

A Declarative System for Optimizing AI Workloads

A long-standing goal of data management systems has been to build systems which can compute quantitative insights over large corpora of unstructured data in a cost-effective manner. Until recently, it was difficult and expensive to extract facts from company documents, data from scientific papers, or metrics from image and video corpora. Today's models can accomplish these tasks with high accuracy. However, a programmer who wants to answer a substantive AI-powered query must orchestrate large numbers of models, prompts, and data operations. For even a single query, the programmer has to make a vast number of decisions such as the choice of model, the right inference method, the most cost-effective inference hardware, the ideal prompt design, and so on. The optimal set of decisions can change as the query changes and as the rapidly-evolving technical landscape shifts. In this paper we present Palimpzest, a system that enables anyone to process AI-powered analytical queries simply by defining them in a declarative language. The system uses its cost optimization framework to implement the query plan with the best trade-offs between runtime, financial cost, and output data quality. We describe the workload of AI-powered analytics tasks, the optimization methods that Palimpzest uses, and the prototype system itself. We evaluate Palimpzest on tasks in Legal Discovery, Real Estate Search, and Medical Schema Matching. We show that even our simple prototype offers a range of appealing plans, including one that is 3.3x faster and 2.9x cheaper than the baseline method, while also offering better data quality. With parallelism enabled, Palimpzest can produce plans with up to a 90.3x speedup at 9.1x lower cost relative to a single-threaded GPT-4 baseline, while obtaining an F1-score within 83.5% of the baseline. These require no additional work by the user.

cs.CL↗

arXiv2023

Accelerating Aggregation Queries on Unstructured Streams of Data

Analysts and scientists are interested in querying streams of video, audio, and text to extract quantitative insights. For example, an urban planner may wish to measure congestion by querying the live feed from a traffic camera. Prior work has used deep neural networks (DNNs) to answer such queries in the batch setting. However, much of this work is not suited for the streaming setting because it requires access to the entire dataset before a query can be submitted or is specific to video. Thus, to the best of our knowledge, no prior work addresses the problem of efficiently answering queries over multiple modalities of streams. In this work we propose InQuest, a system for accelerating aggregation queries on unstructured streams of data with statistical guarantees on query accuracy. InQuest leverages inexpensive approximation models ("proxies") and sampling techniques to limit the execution of an expensive high-precision model (an "oracle") to a subset of the stream. It then uses the oracle predictions to compute an approximate query answer in real-time. We theoretically analyzed InQuest and show that the expected error of its query estimates converges on stationary streams at a rate inversely proportional to the oracle budget. We evaluated our algorithm on six real-world video and text datasets and show that InQuest achieves the same root mean squared error (RMSE) as two streaming baselines with up to 5.0x fewer oracle invocations. We further show that InQuest can achieve up to 1.9x lower RMSE at a fixed number of oracle invocations than a state-of-the-art batch setting algorithm.

cs.DB↗

arXiv2015

Radially Magnetized Protoplanetary Disk: Vertical Profile

This paper studies the response of a thin accretion disk to an external radial magnetic field. Our focus is on protoplanetary disks (PPDs), which are exposed during their later evolution to an intense, magnetized wind from the central star. A radial magnetic field is mixed into a thin surface layer, is wound up by the disk shear, and is pushed downward by a combination of turbulent mixing and ambipolar and Ohmic drift. The toroidal field reaches much greater strengths than the seed vertical field that is usually invoked in PPD models, even becoming superthermal. Linear stability analysis indicates that the disk experiences the magnetorotational instability (MRI) at a higher magnetization than a vertically magnetized disk when both the effects of ambipolar and Hall drift are taken into account. Steady vertical profiles of density and magnetic field are obtained at several radii between 0.06 and 1 AU in response to a wind magnetic field $B_r \sim (10^{-4}$-$10^{-2})(r/{\rm AU})^{-2}$ G. Careful attention is given to the radial and vertical ionization structure resulting from irradiation by stellar X-rays. The disk is more strongly magnetized closer to the star, where it can support a higher rate of mass transfer. As a result, the inner $\sim 1$ AU of a PPD is found to evolve toward lower surface density. Mass transfer rates around $10^{-8}\,M_\odot$ yr$^{-1}$ are obtained under conservative assumptions about the MRI-generated stress. The evolution of the disk, and the implications for planet migration, are investigated in the accompanying paper.

astro-ph.EP↗

arXiv2015

Constrained Evolution of a Radially Magnetized Protoplanetary Disk: Implications for Planetary Migration

We consider the inner $\sim$ AU of a protoplanetary disk (PPD), at a stage where angular momentum transport is driven by the mixing of a radial magnetic field into the disk from a T-Tauri wind. Because the radial profile of the imposed magnetic field is well constrained, a deterministic calculation of the disk mass flow becomes possible. The vertical disk profiles obtained in Paper I imply a stronger magnetization in the inner disk, faster accretion, and a secular depletion of the disk material. Inward transport of solids allows the disk to maintain a broad optical absorption layer even when the grain abundance becomes too small to suppress its ionization. Thus a PPD may show a strong middle-to-near infrared spectral excess even while its mass profile departs radically from the minimum-mass solar nebula. The disk surface density is buffered at $\sim 30$ g cm$^{-2}$: below this, X-rays trigger strong enough magnetorotational turbulence at the midplane to loft mm-cm sized particles high in the disk, followed by catastrophic fragmentation. A sharp density gradient bounds the inner depleted disk, and propagates outward to $\sim 1$-2 AU over a few Myr. Earth-mass planets migrate through the inner disk over a similar timescale, whereas the migration of Jupiters is limited by the supply of gas. Gas-mediated migration must stall outside 0.04 AU, where silicates are sublimated and the disk shifts to a much lower column. A transition disk emerges when the dust/gas ratio in the MRI-active layer falls below $X_d \sim 10^{-6}(a_d/\mu{\rm m})$, where $a_d$ is the grain size.

astro-ph.EP↗

arXiv2014

Integrable Spatiotemporally Varying NLS, PT-Symmetric NLS, and DNLS Equations: Generalized Lax Pairs and Lie Algebras

This paper develops two approaches to Lax-integrbale systems with spatiotemporally varying coefficients. A technique based on extended Lax Pairs is first considered to derive variable-coefficient generalizations of various Lax-integrable NLPDE hierarchies recently introduced in the literature. As illustrative examples, we consider generalizations of the NLS and DNLS equations, as well as a PT-symmetric version of the NLS equation. It is demonstrated that the techniques yield Lax- or S-integrable NLPDEs with both time- AND space-dependent coefficients which are thus more general than almost all cases considered earlier via other methods such as the Painleve Test, Bell Polynomials, and various similarity methods. However, this technique, although operationally effective, has the significant disadvantage that, for any integrable system with spatiotemporally varying coefficients, one must 'guess' a generaliza- tion of the structure of the known Lax Pair for the corresponding system with constant coefficients. Motivated by the somewhat arbitrary nature of the above procedure, we therefore next attempt to systematize the derivation of Lax-integrable sytems with variable coefficients. We attempt to apply the Estabrook- Wahlquist (EW) prolongation technique, a relatively self-consistent procedure requiring little prior infomation. However, this immediately requires that the technique be significantly generalized or broadened in several different ways. The new and extended EW technique which results is illustrated by algorithmically deriving generalized Lax-integrable versions of NLS, PT-symmetric NLS, and DNLS equations.

math.AP↗

arXiv2014

The Extended Estabrook-Wahlquist Method

Variable Coefficient Korteweg de Vries (vcKdV), Modified Korteweg de Vries (vcMKdV), and nonlinear Schrodinger (NLS) equations have a long history dating from their derivation in various applications. A technique based on extended Lax Pairs has been devised recently to derive variable-coefficient generalizations of various Lax-integrable NLPDE hierarchies. The resulting Lax- or S-integrable NLPDEs have both time- AND space-dependent coefficients, and are thus more general than almost all cases considered earlier via other methods such as the Painlev\'e Test, Bell Polynomials, and various similarity methods. However, this technique, although operationally effective, has the significant disadvantage that, for any integrable system with spatiotemporally varying coefficients, one must 'guess' a generalization of the structure of the known Lax Pair for the corresponding system with constant coefficients. Motivated by the somewhat arbitrary nature of the above procedure, we embark in this paper on an attempt to systematize the derivation of Lax-integrable systems with variable coefficients. An ideal approach would be a method which does not require knowledge of the Lax pair to an associated constant coefficient system, and also involves little to no guesswork. Hence we attempt to apply the Estabrook-Wahlquist (EW) prolongation technique, a relatively self-consistent procedure requiring little prior information. The new and extended EW technique which results is illustrated by algorithmically deriving generalized Lax-integrable versions of the NLS, generalized fifth-order KdV, MKdV, and derivative nonlinear Schrodinger (DNLS) equations.

nlin.SI↗

arXiv2014

Integrable Generalized KdV and MKdV Equations with Spatiotemporally Varying Coefficients

A technique based on extended Lax Pairs is first considered to derive variable-coefficient generalizations of various Lax-integrable NLPDE hierarchies recently introduced in the literature. As illustrative examples, we consider generalizations of KdV equations and three variants of generalized MKdV equations. It is demonstrated that the techniques yield Lax- or S-integrable NLPDEs with both time- AND space-dependent coefficients which are thus more general than almost all cases considered earlier via other methods such as the Painleve Test, Bell Polynomials, and various similarity methods. However, this technique, although operationally effective, has the significant disadvantage that, for any integrable system with spatiotemporally varying coefficients, one must guess a generalization of the structure of the known Lax Pair for the corresponding system with constant coefficients. Motivated by the somewhat arbitrary nature of the above procedure, we embark in this paper on an attempt to systematize the derivation of Lax-integrable sytems with variable coefficients. Hence we attempt to apply the Estabrook-Wahlquist (EW) prolongation technique, a relatively self-consistent procedure requiring little prior information. However, this immediately requires that the technique be significantly generalized or broadened in several different ways, including solving matrix partial differential equations instead of algebraic ones. The new and extended EW technique whch results is illustrated by algorithmically deriving generalized Lax-integrable versions of the generalized fifth-order KdV, and MKdV equations.

math-ph↗

arXiv2013

Hot Electromagnetic Outflows I: Acceleration and Spectra

The theory of cold, relativistic, magnetohydrodynamic outflows is generalized by the inclusion of an intense radiation source. In some contexts, such the breakout of a gamma-ray burst jet from a star, the outflow is heated to a high temperature at a large optical depth. Eventually it becomes transparent and is pushed to a higher Lorentz factor by a combination of the Lorentz force and radiation pressure. We obtain its profile, both inside and outside the fast magnetosonic critical point, when the poloidal magnetic field is radial and monopolar. Most of the energy flux is carried by the radiation field and the toroidal magnetic field that is wound up close to the rapidly rotating engine. Although the entrained matter carries little energy, it couples the radiation field to the magnetic field. Then the fast critical point is pushed inward from infinity and, above a critical radiation intensity, the outflow is accelerated mainly by radiation pressure. We identify a distinct observational signature of this hybrid outflow: a hardening of the radiation spectrum above the peak of the seed photon distribution, driven by bulk Compton scattering. The non-thermal spectrum -- obtained by a Monte Carlo method -- is most extended when the Lorentz force dominates the acceleration, and the seed photon beam is wider than the Lorentz cone of the MHD fluid. This effect is a generic feature of hot, magnetized outflows interacting with slower relativistic material. It may explain why some GRB spectra appear to peak at photon energies above the original Amati et al. scaling. A companion paper addresses the case of jet breakout, where diverging magnetic flux surfaces yield strong MHD acceleration over a wider range of Lorentz factor.

astro-ph.HE↗

arXiv2013

Hot Electromagnetic Outflows II: Jet Breakout

We consider the interaction between radiation, matter and a magnetic field in a compact, relativistic jet. The entrained matter accelerates outward as the jet breaks out of a star or other confining medium. In some circumstances, such as gamma-ray bursts (GRBs), the magnetization of the jet is greatly reduced by an advected radiation field while the jet is optically thick to scattering. Where magnetic flux surfaces diverge rapidly, a strong outward Lorentz force develops and radiation and matter begin to decouple. The increase in magnetization is coupled to a rapid growth in Lorentz factor. We take two approaches to this problem. The first examines the flow outside the fast magnetosonic critical surface, and calculates the flow speed and the angular distribution of the radiation field over a range of scattering depths. The second considers the flow structure on both sides of the critical surface in the optically thin regime, using a relaxation method. In both approaches, we find how the terminal Lorentz factor, and radial profile of the outflow, depend on the radiation intensity and optical depth at breakout. The effect of bulk Compton scattering on the radiation spectrum is calculated by a Monte Carlo method, while neglecting the effects of internal dissipation. The peak of the scattered spectrum sits near the seed peak if radiation pressure dominates the acceleration, but is pushed to a higher frequency if the Lorentz force dominates, and especially if the seed photon cone is broadened by interaction with a slower component of the outflow.

astro-ph.HE↗

arXiv2011

Light Loop Echoes and Blinking Black Holes

Radiation emitted near a black hole reaches the observer by multiple paths; and when this radiation varies in time, the time-delays between the various paths generate a "blinking" effect in the observed light curve L(t) or its auto-correlation function xi(T)= . For the particularly important "face-on" configuration (in which the hole is viewed roughly along its spin axis, while the emission comes roughly from its equatorial plane -- e.g. from the inner edge of its accretion disk, or from the violent flash of a nearby/infalling star) we calculate the blinking in detail by computing the time delay Delta t_{j}(r,a) and magnification mu_{j}(r,a) of the jth path (j=1,2,3,...), relative to the primary path (j=0), as a function of the emission radius r and black hole spin 0<a/M<1. The particular geometry and symmetry of the nearly-face-on configuration enhances and "protects" the blinking signal, making it more detectable and more independent of certain astrophysical and observational details. The effect can be surprisingly strong: e.g. for radiation from the innermost stable circular orbit ("ISCO") of a black hole of critical spin (a_{crit}/M = 0.853), the j=1,2,3 fluxes are, respectively, 27%, 2% and 0.1% of the j=0 flux.

astro-ph.HE↗