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

Relational database management systems (RDBMSes) can process general-purpose queries, but often have lower performance compared to custom-built solutions for specific queries. For example, consider a group-by query over a few known groups (e.g., grouping by country). While an RDBMS would likely use a hash map to do the grouping, a faster method could hard-code the expected groups into the query executor. But such workload-specific techniques, which we call query accelerators, are not widely used in practice because the engineering effort (optimizer and engine changes, potential bugs) does not justify the isolated performance gains (speedup on a single specific query). We propose Tailwind: an external query planner that brings accelerators into any RDBMS that supports data import/export. Users define their accelerators using abstract logical plans (ALPs): a new mostly-declarative abstraction over relational operators built on regular tree expressions. ALPs allow Tailwind to automatically build customized neural network models to estimate when using a particular accelerator is beneficial. At runtime, Tailwind sits atop an RDBMS and transparently rewrites queries to run across one or more accelerators when predicted to be beneficial, falling back to the underlying RDBMS when not. On Redshift and DuckDB with a library of four diverse accelerators, Tailwind accelerates TPC-H queries by 1.38x on average (up to 29x).

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.

Recent work has shown that LLMs can synthesize highly specialized database systems for fixed workloads, but existing approaches typically assume static data and replace the database's native storage with generated representations. We present JetStream, a system for generating query-specific accelerators that instead extend an existing DBMS. JetStream consists of three parts. First, a staged, measurement-driven agentic workflow generates and optimizes query-specific accelerators, including persistent auxiliary state when beneficial. Second, a fixed, engine-neutral substrate provides the common interfaces for execution, transaction coordination, state management, and maintenance. Third, a separate synthesis workflow generates engine-specific backend adapters that connect the substrate to the underlying engine. For stateful accelerators, JetStream also generates maintenance logic and uses a runtime policy to choose among incremental maintenance, rebuilds, and lazy repair as the database changes. On TPC-H at SF=20, stateful accelerators generated by JetStream achieve an 833x geomean read-only speedup over DuckDB, compared with 34.07x for GenDB and 12.35x for Bespoke OLAP. Under TPC-H refreshes every 60 seconds, JetStream maintains a 375x geomean workload speedup. We also show that JetStream generalizes to new, unseen workloads, achieving geomean read-only speedups of 102x over DuckDB and 486x over PostgreSQL. These results show that aggressive generated specialization can be integrated seamlessly with existing DBMSes and support dynamic workloads.

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. Evaluating 8 LLMs with our single-agent reference framework DS-Guru, alongside open- and closed-source agentic systems, we find that while current single-agent 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 50% end-to-end accuracy in the full data-lake setting. Even with perfect retrieval, the accuracy tops out at 59%. Leading LLMs can identify up to 42% of important data tasks but can only fully implement 20% of individual data tasks.

Modern cloud data warehouses decouple compute from storage, making it easy for organizations to access the same underlying data with multiple compute clusters. This flexibility is often used for performance isolation among diverse workloads, so that each workload meets its latency service-level objective (SLO) more reliably. For example, interactive dashboards, ad hoc analysis, and batch jobs can each run on separate clusters. However, this dedicated-cluster approach requires each compute cluster to be continuously scaled to adapt to workload evolution, with over-provisioning wasting resources and under-provisioning risking SLO violations. We present AutoSLO, a latency-SLO-aware workload management framework for multi-cluster cloud data warehouses. AutoSLO operates across three timescales through three key components. First, a periodic Policy Tuner plans proactive cluster scaling actions and tunes configuration parameters, using simulations of history-derived workload forecasts. Second, an SLO-aware reactive Autoscaler adjusts the active cluster set when recent workload behavior deviates from the forecast. Third, an online Query Router reacts to live load when placing each query, using a concurrency-aware latency predictor to avoid SLO violations. On realistic Redbench workloads, AutoSLO successfully meets latency SLOs of varying strictness, reducing cost by a mean of 26.4

Query planners are typically expected to produce optimized plans quickly, leading many researchers (including the authors of this paper) and practitioners to design systems that assume query planning is a low-cost operation. Using a lightweight agentic search, we show that this assumption does not always hold. Across seven DBMSes, including four commercial systems, we find at least one query per system that takes more than three minutes to plan. In addition to being slow to plan, such queries risk tying up database resources without performing useful work, creating a potential denial-of-service vector. We analyze the queries our search uncovers and compare how the seven systems respond to each pattern. We find that although the queries triggering slow planning are largely DBMS-specific, recurring pathologies involving correlated subqueries, CTE expansion, repeated subquery expressions, disjunctive joins, and constant folding affect multiple systems. We release our uncovered queries along with a curated suite of parameterized query pathologies that researchers and database engineers can use to test planner robustness. Overall, our results show that query planning cannot always be treated as a predictably inexpensive operation and that its latency and robustness deserve further attention from both database researchers and engineers.

Recent advances in LLM-based Text-to-SQL have achieved remarkable gains on public benchmarks such as BIRD and Spider. Yet, these systems struggle to scale in realistic enterprise settings with large, complex schemas, diverse SQL dialects, and expensive multi-step reasoning. Emerging agentic approaches show potential for adaptive reasoning but often suffer from inefficiency and instability-repeating interactions with databases, producing inconsistent outputs, and occasionally failing to generate valid answers. To address these challenges, we introduce Agent Semantic Memory (AgentSM), an agentic framework for Text-to-SQL that builds and leverages interpretable semantic memory. Instead of relying on raw scratchpads or vector retrieval, AgentSM captures prior execution traces-or synthesizes curated ones-as structured programs that directly guide future reasoning. This design enables systematic reuse of reasoning paths, which allows agents to scale to larger schemas, more complex questions, and longer trajectories efficiently and reliably. Compared to state-of-the-art systems, AgentSM achieves higher efficiency by reducing average token usage and trajectory length by 25

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

NL2SQL (natural language to SQL) systems translate natural language into SQL queries, allowing users with no technical background to interact with databases and create tools like reports or visualizations. While recent advancements in large language models (LLMs) have significantly improved NL2SQL accuracy, schema ambiguity remains a major challenge in enterprise environments with complex schemas, where multiple tables and columns with semantically similar names often co-exist. To address schema ambiguity, we introduce ODIN, a NL2SQL recommendation engine. Instead of producing a single SQL query given a natural language question, ODIN generates a set of potential SQL queries by accounting for different interpretations of ambiguous schema components. ODIN dynamically adjusts the number of suggestions based on the level of ambiguity, and ODIN learns from user feedback to personalize future SQL query recommendations. Our evaluation shows that ODIN improves the likelihood of generating the correct SQL query by 1.5-2$\times$ compared to baselines.

In the classical problem formulation of Text-to-SQL in academia, each question is translated independently of the others into SQL. This differs from the setting in practice, where questions enter the Text-to-SQL system in sequence. Thus, for all but the first few questions, a translation history is available that contains past questions and how they were translated by the system. So far, it has not been sufficiently explored how Text-to-SQL systems can make use of the translation history to improve future translations. Another crucial difference from the academic setting is that in practice, users generally have a conversation with the Text-to-SQL system. More concretely, if the initial translation contains a mistake, the user can follow up with feedback messages to allow the system to fix the mistake. In this case, it might be helpful to remember these user feedback messages to avoid repeating past mistakes. Thus, in this paper, we explore how a history of such past conversations between users and the Text-to-SQL system can be used to make future Text-to-SQL translations more accurate. We explore several approaches for extracting relevant experiences and insights from this conversation history and show in an evaluation that utilizing them can improve translation accuracy by up to 14.9%.

On October 19 and 20, 2023, the authors of this report convened in Cambridge, MA, to discuss the state of the database research field, its recent accomplishments and ongoing challenges, and future directions for research and community engagement. This gathering continues a long standing tradition in the database community, dating back to the late 1980s, in which researchers meet roughly every five years to produce a forward looking report. This report summarizes the key takeaways from our discussions. We begin with a retrospective on the academic, open source, and commercial successes of the community over the past five years. We then turn to future opportunities, with a focus on core data systems, particularly in the context of cloud computing and emerging hardware, as well as on the growing impact of data science, data governance, and generative AI. This document is not intended as an exhaustive survey of all technical challenges or industry innovations in the field. Rather, it reflects the perspectives of senior community members on the most pressing challenges and promising opportunities ahead.

Constructing real-world data-to-insight pipelines often involves data extraction from data lakes, data integration across heterogeneous data sources, and diverse operations from data cleaning to analysis. The design and implementation of data science pipelines require domain knowledge, technical expertise, and even project-specific insights. AI systems have shown remarkable reasoning, coding, and understanding capabilities. However, it remains unclear to what extent these capabilities translate into successful design and execution of such complex pipelines. We introduce KRAMABENCH: a benchmark composed of 104 manually-curated real-world data science pipelines spanning 1700 data files from 24 data sources in 6 different domains. We show that these pipelines test the end-to-end capabilities of AI systems on data processing, requiring data discovery, wrangling and cleaning, efficient processing, statistical reasoning, and orchestrating data processing steps given a high-level task. Our evaluation tests 5 general models and 3 code generation models using our reference framework, DS-GURU, which instructs the AI model to decompose a question into a sequence of subtasks, reason through each step, and synthesize Python code that implements the proposed design. Our results on KRAMABENCH show that, although the models are sufficiently capable of solving well-specified data science code generation tasks, when extensive data processing and domain knowledge are required to construct real-world data science pipelines, existing out-of-box models fall short. Progress on KramaBench represents crucial steps towards developing autonomous data science agents for real-world applications. Our code, reference framework, and data are available at https://github.com/mitdbg/KramaBench.

Sideways information passing is a well-known technique for mitigating the impact of large build sides in a database query plan. As currently implemented in production systems, sideways information passing enables only a uni-directional information flow, as opposed to instance-optimal algorithms, such as Yannakakis'. On the other hand, the latter require an additional pass over the input, which hinders adoption in production systems. In this paper, we make a step towards enabling single-pass bi-directional information passing during query execution. We achieve this by statically analyzing between which tables the information flow is blocked and by leveraging precomputed join-induced fingerprint columns on FK-tables. On the JOB benchmark, Parachute improves DuckDB v1.2's end-to-end execution time without and with semi-join filtering by 1.54x and 1.24x, respectively, when allowed to use 15% extra space.

Query scheduling is a critical task that directly impacts query performance in database management systems (DBMS). Deeply integrated schedulers, which require changes to DBMS internals, are usually customized for a specific engine and can take months to implement. In contrast, non-intrusive schedulers make coarse-grained decisions, such as controlling query admission and re-ordering query execution, without requiring modifications to DBMS internals. They require much less engineering effort and can be applied across a wide range of DBMS engines, offering immediate benefits to end users. However, most existing non-intrusive scheduling systems rely on simplified cost models and heuristics that cannot accurately model query interactions under concurrency and different system states, possibly leading to suboptimal scheduling decisions. This work introduces IconqSched, a new, principled non-intrusive scheduler that optimizes the execution order and timing of queries to enhance total end-to-end runtime as experienced by the user - query queuing time plus system runtime. Unlike previous approaches, IconqSched features a novel predictor, Iconq, which treats the DBMS as a black box and accurately estimates the system runtime of concurrently executed queries under different system states. Using these predictions, IconqSched is able to capture system runtime variations across different query mixes and system loads. It then employs a greedy scheduling algorithm to effectively determine which queries to submit and when to submit them. We compare IconqSched to other schedulers in terms of end-to-end runtime using realistic workload traces. On Postgres, IconqSched reduces end-to-end runtime by up to 16.5% on average and 33.6% in the tail. Similarly, on Redshift, it reduces end-to-end runtime by up to 14.4% on average and 22.9% in the tail.

Cloud service providers commonly use standard benchmarks like TPC-H and TPC-DS to evaluate and optimize cloud data analytics systems. However, these benchmarks rely on fixed query patterns and fail to capture the real execution statistics of production cloud workloads. Although some cloud database vendors have recently released real workload traces, these traces alone do not qualify as benchmarks, as they typically lack essential components like the original SQL queries and their underlying databases. To overcome this limitation, this paper introduces a new problem of workload synthesis with real statistics, which aims to generate synthetic workloads that closely approximate real execution statistics, including key performance metrics and operator distributions, in real cloud workloads. To address this problem, we propose PBench, a novel workload synthesizer that constructs synthetic workloads by judiciously selecting and combining workload components (i.e., queries and databases) from existing benchmarks. This paper studies the key challenges in PBench. First, we address the challenge of balancing performance metrics and operator distributions by introducing a multi-objective optimization-based component selection method. Second, to capture the temporal dynamics of real workloads, we design a timestamp assignment method that progressively refines workload timestamps. Third, to handle the disparity between the original workload and the candidate workload, we propose a component augmentation approach that leverages large language models (LLMs) to generate additional workload components while maintaining statistical fidelity. We evaluate PBench on real cloud workload traces, demonstrating that it reduces approximation error by up to 6x compared to state-of-the-art methods.

We study allowing large language models (LLMs) to process arbitrarily long prompts through the lens of inference-time scaling. We propose Recursive Language Models (RLMs), a general inference strategy that treats long prompts as part of an external environment and allows the LLM to programmatically examine, decompose, and recursively call itself over snippets of the prompt. We find that RLMs successfully handle inputs up to two orders of magnitude beyond model context windows and, even for shorter prompts, dramatically outperform the quality of base LLMs and common long-context scaffolds across four diverse long-context tasks, while having comparable (or cheaper) cost per query.