enterprise search engineers

Abstract: Semantic filtering selects records that satisfy a natural-language predicate and is an important operation in data analytics based on large language models (LLMs). Executing an expensive model on every document incurs substantial inference cost. Existing approaches reduce this cost by accepting predictions from cheaper models and forwarding uncertain records to an oracle. Task Cascades further varies document portions and operations, and provides an accuracy-guaranteed extension. However, its threshold calibration follows a prescribed adjustment sequence after cascade assembly, rather than minimizing cost over alternative combinations of stage thresholds. Motivated by this, we propose CascadeFilter, a three-stage semantic filtering method that jointly searches the thresholds for a small language model operating on extracted evidence and full documents, with unresolved records forwarded to the oracle. For a fixed cascade, we formulate threshold selection as minimizing token-aware cost subject to an oracle-relative error constraint, and select the least-cost certified configuration in a pre-specified grid. A finite-population calibration procedure bounds oracle-relative error after selection, using fixed-sequence tests on threshold tracks to allocate the multiple-testing budget. We further combine lexical and dense retrieval, reranking, and neighborhood-supported confidence to improve evidence extraction and routing. These components enhance the cascade while leaving its certification interface unchanged. Across six datasets and 13 evaluation settings, the reported results have the highest benchmark accuracy and macro-averaged F1 score in 12 settings each. These comparisons describe the reported operating points; the accuracy guarantee concerns agreement with the oracle.

Abstract: Retrieval-Augmented Generation (RAG) for multi-hop reasoning requires retrieval to adapt as intermediate evidence reveals new information needs. Existing approaches support such retrieval through explicit graph structures or iterative query reformulation, introducing graph-construction or query-time overhead. We present Evidence-Guided Adaptive Graph Retrieval (EAGR), a framework in which accumulated evidence dynamically determines which semantic directions to explore next, and which requires no dedicated graph database and no explicit graph materialization. EAGR operates over an implicit semantic graph induced by compact passage metadata pre-ingested into a standard VectorDB. Starting from dense-retrieval seed passages, an LLM iteratively assesses accumulated evidence, identifies promising semantic directions, expands the retrieval frontier through metadata-based neighborhood lookup, and terminates once sufficient evidence has been collected. By precomputing semantic metadata, EAGR enables evidence-conditioned graph exploration without materializing an explicit graph, maintaining a dedicated GraphDB, or iteratively rewriting the query. Across three widely used multi-hop QA benchmarks, EAGR achieves the highest LLM-based answer accuracy among evaluated methods, outperforming the vanilla RAG by 17\% and strongest graph-aware baseline by 4\% percentage points averaged across three popular benchmarks HotPotQA, 2WikiMultiHopQA, and MuSiQue.

Abstract: While language models have become dramatically more capable, retrieval still largely relies on lexical matching or similarity in fixed embedding spaces. This creates a capability asymmetry: what the retriever fails to find, even the strongest generator cannot recover. We propose DecisionRAG, which brings the capabilities of pretrained language models to retrieval using decision models that jointly evaluate candidate branches in a single forward pass, without generating any tokens, and show that judging candidates in a shared context makes their relevance estimates comparable. Like a person browsing a library, DecisionRAG navigates the corpus's inherent structure from broad topics down to individual passages, deciding at each level which branches are worth exploring. Without task-specific training or a first-stage retriever, DecisionRAG outperforms dense retrieval with an 8B reranker by 13.8 R@1 points on MultiDoc2Dial and 17.4 Hit@1 points on Natural Questions. Across four benchmarks, including a corpus of 2.68 million Wikipedia paragraphs, retrieval takes just 0.7-1.8 seconds per query.

Abstract: Dense retrievers are commonly trained with hard negatives selected from lexical or dense retrieval results. Retrieval proximity, however, does not indicate whether a candidate conflicts with explicit domain structure. We introduce CA-HNM, a constraint-aware hard-negative mining framework that combines retrieval-based candidate generation with deterministic incompatibility rules derived from a structured domain model. The framework removes known positives, constructs a local typed context around the query target, records relational, scope, target, and level mismatches, and uses the resulting candidates to fine-tune a standard dual encoder. Structured information is required only during offline negative construction. We evaluate CA-HNM on MOOCCubeX and Course-Skill Atlas with two encoders, two training losses, and ten seeds. Relative to a retrieval-rank-matched control, CA-HNM-Pure has higher mean NDCG@10 in six of eight dataset--encoder--loss configurations. Four hierarchical 95% bootstrap intervals are entirely above zero, mostly under triplet loss. In a separate experiment, CA-HNM-Mixed shows slightly higher mean NDCG@10 than a matched-mixed control on both collections. Structural variants show that changes in available structure can also change candidate eligibility and negative-pool composition. The results indicate that the contribution of constraint-aware selection depends on the training objective and the composition of the negative pool.

Abstract: Industrial retrieval systems often improve quality by enlarging embedding candidate pools or adopting complex training pipelines. We show that, under a sufficiently large but practical candidate pool, embedding recall exhibits diminishing returns, motivating a closer examination of reranker quality. On Chinese legal retrieval benchmarks, embedding top-300 reaches a high-recall operating region, while relevant evidence may still be ranked below the output cutoff. We compare hard-negative strategies for reranker training, including embedding-mined, self-mined, and dynamic-negative variants, under aligned retrieval protocols that evaluate reranking after candidate retrieval. Among the evaluated staged configurations, an embedding-to-self-mined pipeline provides a favorable trade-off; additional dynamic stages do not improve all held-out metrics. The selected pipeline scales from a 0.6B to a 4B reranker. Relative to the native 4B reranker, the trained 4B Stage2 model improves reported metrics on the STARD and CSAID benchmarks and on an internal development set, and is preferred in 72.4\% of non-tied comparisons on 410 de-identified, reconstructed internal operational-query inputs in a latency-sensitive offline preference evaluation. These results support evaluating the retrieval and reranking stages separately when selecting an optimization target for industrial retrieval systems.

Abstract: High-precision retrieval over external knowledge bases is essential for LLM agents to operate effectively. However, existing Information Retrieval (IR) research primarily optimizes query-side retrievers (e.g., ColBERT, BGE) and generators (e.g., Self-RAG, GraphRAG), while reducing corpus-side representation to chunking. This leaves index-side quality an underexplored lever. We propose Pre-Retrieval Index Modeling & Enhancement (PRIME), an index-side optimization framework built on two complementary tracks: an LLM-driven Semantic Restructuring (SR) track that transforms long documents into structured, information-dense sections, and a deterministic Metadata Synthesis (MS) and Hierarchical Retrieval (HR) track that extracts semantic anchors and heading structure for coarse-to-fine section retrieval. Our evaluations show that deterministic metadata matches structure-aware chunking's retrieval quality while cutting embedding computation by 45.8 $\times$ and index vector count by 22.4 $\times$. On top of these gains, Semantic Restructuring (SR) further improves Answer-F1 by 13.7% in full deployment, with characterized quality-fidelity-cost trade-offs. Across QASPER, BRIGHT, and NFCorpus, we further characterize three qualitative regimes that determine whether index-side optimization helps or degrades retrieval, while showing that these index-side effects are neutralized under a strong reranker.

Abstract: Memory-augmented assistants retrieve evidence from stored conversations and documents to answer queries. Most embedding-based retrievers compare single-vector representations of queries and memories, even when relevance depends on a small part of the content. Such local evidence can be matched to queries using the contextual token representations already computed by the same embedding models. Our method uses sparse token matching to improve retrieval without additional training. Each query token retrieves a small set of matching tokens across candidate memories. The matching scores are aggregated for each memory and combined with whole-memory similarity to rank memories. An optional low-rank query adapter can be trained with supervision while keeping the encoder and stored memory representations fixed. Our method improves retrieval across eight benchmarks and multiple model families without additional training. Query adaptation further increases the gains to 13.3 and 14.3 percentage points in average Recall@10 and NDCG@10 on Qwen3-Embedding-4B. These retrieval gains translate into higher question-answering accuracy on long-term memory benchmarks.

Abstract: Search agents enable Large Language Models (LLMs) to iteratively retrieve and use information for complex multi-hop questions. Reinforcement Learning with Verifiable Rewards (RLVR) offers a promising approach for post-training such agents, but its reliance on sparse, outcome-based supervision can make credit assignment difficult and limit learning efficiency. In this paper, we systematically investigate how intermediate supervision can improve reinforcement learning for search agents. We study a range of reward-shaping and credit-assignment strategies that provide learning signals from intermediate retrieval steps. Building on these insights, we develop a training framework that combines intermediate signals with final outcome rewards to improve learning from multi-step search trajectories. Experiments across multiple benchmarks under matched training conditions demonstrate improvements in aggregate search-agent performance and show that both the choice of intermediate signal and where its credit is assigned affect training behaviour. These findings show that reward design and credit assignment are important design dimensions for training effective search agents.

Abstract: Continuous ingestion can expose new retrieval-augmented generation (RAG) content to retrieval before vetting completes, creating a temporal attack surface that conventional admission decisions do not capture. We present a fail-closed provisional-visibility protocol that admits new content under a deadline, hides items whose verification has not committed in time, and supports lineage-scoped containment. The protocol bounds unvetted exposure by the configured visibility budget plus query-path enforcement delay; poisoning exposure remains conditional on verifier correctness. We evaluate the design on five workloads of encoded natural-language documents using an exact backend and Milvus. Under verifier backlog, the deadline reduced median poisoned retrievals from 34 (29--34) to 7 (6--7) relative to asynchronous admission without a deadline, with the same reduction in displaced clean results. Verify-before-visible avoided provisional exposure but delayed clean first visibility to 6.3 s, whereas provisional admission made content visible within 3 ms; expired clean items could nevertheless experience temporary availability gaps. Replaying decisions from a recipe-specific detector illustrated the protocol boundary: false promotions left poison visible, while false refusals excluded clean content from retrieval. Standalone Milvus tests characterized enforcement delay and concurrent HNSW retrieval on one node. These results clarify which guarantees come from the protocol and which outcomes depend on detector quality, while quantifying the freshness and availability trade-offs.

Abstract: Adaptive RAG uses signals such as confidence, relevance, support, and retrieval quality to decide when to search or correct evidence. In multi-step retrieval, however, these local signals must be combined into a persistent view of what the current evidence supports, what remains missing, and which action should follow. Existing methods often use such signals as separate triggers, making it difficult to preserve a coherent evidence state across a trajectory; we call this problem evidence-state fragmentation. We introduce BELIEFRAG, a closed-loop controller that updates an explicit state over sufficiency, reliability, conflict, uncertainty, evidence gaps, and acquisition cost, then chooses among retrieval, query rewriting, verification, answering, stopping, and abstention. Across six QA benchmarks with gpt-oss-120b, BELIEFRAG reaches mean token F1 0.572 with 3.89k tokens per question, outperforming fixed iterative retrieval (0.555 F1) while using 39% fewer tokens. The same quality-cost pattern transfers to Qwen3-32B, where BELIEFRAG reaches 0.552 F1 versus 0.523 for iterative retrieval while using 35% fewer tokens. Analysis shows that the main gains come from corrective re-retrieval rather than pruning alone, while several belief dimensions are redundant and calibrated answerability plays the strongest operational role. Calibration improves threshold stability across related evidence sources, although source shift can still invalidate the same decision signal.

Abstract: Retrieval-Augmented Generation (RAG) systems over enterprise knowledge bases must ingest heterogeneous document formats -- PDFs, Word documents, presentations, and scans -- whose content is locked inside complex visual layouts, multi-column pages, and dense tables. Rule-based extraction and OCR destroy reading order, flatten tables, and lose heading hierarchy, while fully agentic chunking over extracted text incurs high token costs and hallucination risk. We present Document Retrieval-Aware Chunking (D-RAC), an extension of our Web Retrieval-Aware Chunking (W-RAC) framework to arbitrary document formats. D-RAC first normalizes any input document into PDF, exploiting the fact that virtually every format has a faithful, deterministic PDF rendering. A single multimodal LLM pass then converts rendered pages into retrieval-optimized Markdown -- rewriting tables as self-contained prose statements and preserving heading hierarchy -- after which chunking proceeds exactly as in W-RAC: deterministic parsing into ID-addressable units followed by lightweight LLM-based chunk planning over identifiers rather than text. Source text is never regenerated during chunking, preserving W-RAC's cost, determinism, and observability benefits while unlocking every renderable format as a first-class input. On the 236-document, 795-page PDF subset of the RAG-Multi-Corpus benchmark spanning five enterprise domains, D-RAC converts and chunks the entire corpus in 72 minutes with zero errors, producing 1,748 retrieval-ready chunks. Compared to agentic chunking with frontier LLMs, D-RAC reduces chunking-stage output tokens by 95.7%, cutting chunking cost by 77.8% (GPT-4.1 pricing) to 85.6% (Gemini 2.5 Pro pricing) and chunking time by 75%. D-RAC scales linearly to documents of 500+ pages.

Abstract: Recent visual document retrieval (VDR) systems such as ColPali use multi-vector page embeddings, in which patch-level vectors enable fine-grained evidence matching but incur substantial index storage and MaxSim scoring overhead. Post-hoc merging offers a practical route to efficient VDR by reducing this cost without retraining the retriever, but its uniform reconstruction objectives are poorly aligned with the sparse, non-uniform patch usage induced by late-interaction retrieval. Under aggressive compression, this misalignment can preserve rarely used patches while concentrating retrieval activity on too few retained representatives. To address this misalignment, we propose Marginal-Guided Compression with Optimal Transport (MAGIC), a training-free post-hoc compressor for efficient retrieval with frozen multi-vector embeddings. MAGIC derives a MaxSim-induced compression surrogate and optimizes it through a two-marginal entropic optimal-transport formulation, where a retrieval-demand source marginal prioritizes high-use patches and a balanced target marginal regularizes retained-facet usage. Across ViDoRe benchmarks, keep ratios, and retrieval backbones, MAGIC consistently outperforms strong post-hoc compressors, with particularly large gains in the aggressive-compression regime; component ablations verify the complementary effects of its two marginals. We release the code at: https://github.com/xandery-geek/MAGIC.

Abstract: RAG systems are routinely evaluated using synthetic question sets generated from the target document corpus. While this practice provides a useful check on overall retrieval capability, relying exclusively on synthetic benchmarks can mislead under distribution shift and overstate deployment readiness. Synthetic generation spreads questions evenly across the corpus, formulating long, detailed queries; real users put most of their traffic on a few administrative and procedural topics in short queries, while also asking about matters the generator never covers at all. We demonstrate this gap on a university faculty information system, comparing 1,851 synthetic questions generated via Gemini Notebook against 322 authentic queries collected via a student survey. The synthetic and authentic query sets differ significantly: authentic queries average 6.8 words versus 15.7 for the synthetic ones, and draw from only 53 unique sources compared to 165. Consequently, configurations that appear highly effective on synthetic benchmarks experience a substantial performance drop on authentic queries. Importantly, optimizing on synthetic queries selected a higher-latency hybrid retriever. In our setting the sparse retrieval component benefited long synthetic questions but not short authentic ones, costing up to $8\times$ the latency of the fastest configuration we tested. We propose treating synthetic and authentic query sets as complementary extremes of the query-quality spectrum: synthetic data verifies maximum retrieval capacity under idealized conditions, while authentic queries test system robustness to the imprecise, underspecified inputs of real users.

Abstract: RAG systems commonly retrieve a fixed number of documents (top-k) to ground generation, but this static approach is brittle: simple queries suffer over-retrieval (adding noise and cost) while complex queries are under-retrieved, causing recall failures that cascade into incorrect answers. Motivated by the question of how many documents must be retrieved to answer an arbitrary query reliably, we propose a practical, general framework for query-adaptive retrieval depth. Offline, we estimate per-query retrieval difficulty by measuring NDCG under the default retriever and deriving a query-specific "saturation" point k* from the NDCG-k curve. Because computing these signals online is expensive, we cluster a large set of queries in embedding space and summarize each cluster with a recommended retrieval depth that targets high coverage (e.g., \textasciitilde{}95\%) using a mean-plus-variance rule. At runtime, the system assigns an incoming query to a cluster and selects the corresponding top-k in constant time. Compared with post-retrieval confidence methods that rely on clustering retrieved documents, our approach is pre-retrieval and query-centric, making it robust in heterogeneous, case-like corpora and applicable across domains such as legal, healthcare, finance, and enterprise search. Finally, this framework has been tested in full-traffic queries that improved $F_1$ by over 36\% while reducing token usage by 14\% on low-complexity clusters without accuracy loss.

Abstract: Retrieval-Augmented Generation (RAG) has empowered Large Language Models (LLMs) to tackle knowledge-intensive tasks. However, navigating global, heterogeneous knowledge bases (large-scale knowledge graphs and text corpora) for complex reasoning remains a challenge. Existing methods typically employ reactive, graph-driven exploration strategies, which blindly follow graph topology without adapting to the question context or evolving exploration progress, and lack deep bidirectional synergy between graph and text. To address these limitations, we propose CoG (Cognition on Graph), a cognitive-inspired, training-free framework for adaptive knowledge exploration. Drawing inspiration from human problem-solving, CoG performs a continuous plan-explore-reflect cycle, where it proactively formulates investigation plans, performs dual-source retrieval, and dynamically reflects on progress to adjust strategies. Crucially, it establishes deep bidirectional synergy between structured graph and unstructured text, where entities extracted from text dynamically guide graph exploration to bridge knowledge gaps. Extensive experiments on seven multi-hop QA benchmarks demonstrate that CoG significantly outperforms state-of-the-art methods while achieving superior exploration efficiency. Our code and datasets are available at https://github.com/zhougengxian/CoG.

Abstract: Retrieval-augmented generation (RAG) critically depends on retrieving the evidence necessary for effective reasoning. However, this remains particularly challenging in multi-hop question answering (QA), where supporting passages are often linked through intermediate entities and relations that must be progressively uncovered. Existing retrieval approaches typically rely on a single retrieval intent or one-shot query expansion, limiting their ability to adapt to newly retrieved evidence and potentially introducing noisy or redundant retrieval signals. To address these limitations, we propose a training-free multi-hop retrieval framework that integrates evidence-conditioned exploration, passage-specific contrastive refinement, and coverage-aware final ranking. During offline indexing, the framework constructs passage-specific contrastive facets that characterize each passage relative to its semantically similar neighbors, providing fine-grained signals to distinguish closely related candidates. At inference time, the framework iteratively retrieves evidence, generates probes targeting unresolved information needs, refines candidate relevance using the contrastive facets, and selects a complementary set of passages that collectively cover diverse evidence-seeking intents. Experiments on MuSiQue, HotpotQA, and 2WikiMultihopQA demonstrate consistent improvements in retrieval quality and downstream QA performance over baselines.