Mert Can Cakmak

Census-style address records contain heterogeneous structures that must be decomposed into fine-grained fields before they can support linkage, geocoding, and administrative processing. This paper describes the design and implementation of a three-stage Planner–Manager–Worker pipeline using a locally deployed open-weight language model. The prototype includes address-specific extraction, same-model semantic consistency review, formatting, record-preserving chunking, and an exploratory bounded retry pathway. Although the prototype also implements workers for names, phone numbers, email addresses, Social Security numbers, and dates of birth, the quantitative evaluation is limited to address-component parsing. We conduct a controlled pilot using 700 author-generated synthetic records distributed across seven address categories. Each record was first created as a structured object and then rendered into text; the original structured object served as the evaluation reference. Across three repeated executions on the same fixed benchmark, the complete hierarchical configuration obtained a mean micro-averaged address-component exact-match score of 95.7%, compared with 80.9% for the evaluated monolithic LLM reference; the deterministic rule-based reference obtained 52.3%. The fixed prompts were not matched, and realized inference calls, dynamic token allocation, original batching, and reported-run segmentation could not be verified from retained artifacts. No component ablations were conducted. Consequently, the numerical difference cannot be attributed uniquely to planning, task coordination, specialization, semantic review, feedback, chunking, or multi-agent organization. The reference labels were not independently annotated, the benchmark does not estimate performance on naturally occurring records, and the retry pathway was invoked only three times. The contribution is therefore the specification and controlled pilot characterization of an on-premise address-parsing pipeline, rather than evidence of architectural causality, production readiness, broad PII-extraction accuracy, or real-record robustness.

Introduction: Entity resolution (ER) systems often produce different clustering outcomes when applied to the same dataset, especially when parameters, algorithms, or system configurations change. However, in many real-world settings, the true linking structure is unknown, making traditional accuracy-based evaluation difficult. Methods: This paper presents the Case Count Metric System (CCMS), a process and software system for comparing two cluster ER outcomes without requiring a truth set. CCMS classifies how each cluster from the first ER process is transformed by the second process into four mutually exclusive cases: unchanged, merged, partitioned, or overlapping. Results: CCMS produces aggregate case counts, singleton summaries, and per-cluster transformation details to support diagnostic analysis. Example applications using synthetic demographic data and an industrial materials dataset show that CCMS can identify how clustering outcomes change under parameter adjustments and alternative ER systems. Discussion: CCMS provides a practical and interpretable method for comparing ER clustering results when labeled ground truth is unavailable. By distinguishing between over-linking, under-linking, and more complex cluster reorganizations, CCMS offers more actionable insight than single-value similarity measures and supports both research analysis and operational ER evaluation.

Industrial product documentation, including catalogs, datasheets, selection guides, and specification sheets, contains essential technical knowledge for procurement, engineering design, maintenance, and supply-chain operations. Much of this information, however, remains locked in PDFs designed for human reading rather than machine interpretation. This survey reviews the end-to-end pipeline for transforming industrial documents into queryable product knowledge bases. We synthesize 189 works published between 2015 and 2025 across document parsing, information extraction, knowledge base engineering, retrieval-augmented querying, and local deployment. The review shows that although OCR systems, document vision-language models, table parsers, vector databases, and RAG frameworks have matured, their integration for industrial product documentation remains weakly validated. We identify a maturity inversion: several components are deployable or near-deployable, whereas end-to-end industrial pipelines remain at an early readiness level. The survey highlights seven critical gaps involving industrial benchmarks, end-to-end evaluation, dense table extraction, cross-manufacturer schema variation, specification value parsing, multi-path retrieval, and local deployment validation. We argue that the central challenge is the engineering of trustworthy product knowledge bases through schema-aware extraction, value normalization, provenance-preserving storage, data-quality control, and hybrid query routing across SQL, lexical, vector, and graph retrieval. The paper concludes with a prioritized research agenda for reliable, auditable, and maintainable industrial product knowledge infrastructure.

Census-style records contain sensitive personal information that must be transformed into structured fields before it can support record linkage, geocoding, duplicate detection, identity verification, and statistical processing. This task is challenging because addresses appear in many forms, including standard street addresses, apartment and unit addresses, university addresses, military APO/FPO/DPO addresses, rural routes, highway addresses, and attention-line records. Rule-based parsers are efficient and transparent but often fail on non-standard formats. Single-prompt Large Language Model (LLM) approaches improve generalization but can suffer from record skipping, field conflation, and long-context degradation when processing heterogeneous documents. In addition, privacy and governance requirements limit the use of cloud-hosted models for sensitive census-style data. This paper presents a three-stage hierarchical multi-agent architecture for structured knowledge extraction from census-style records using a locally deployed open-weight LLM. A Planner Agent analyzes the input document and formulates an extraction strategy. A Manager Agent converts this strategy into a dependency-aware task graph. A fleet of eight specialized Worker Agents performs extraction, validation, and formatting, while a bounded feedback loop supports limited autonomous recovery from extraction failures. The system runs on gpt-oss-20b, a 21-billion-parameter open-weight model deployed on local infrastructure, so input records do not need to be transmitted to an external model provider. The system is evaluated on 700 synthetically generated records across seven address categories. It achieves 95.7% component-level exact match accuracy, compared with 52.3% for a rule-based baseline and 80.9% for a single-prompt LLM baseline using the same model. The largest improvements occur on challenging non-standard categories, including highway addresses (93% vs. 11% rule-based), military addresses (91% vs. 28%), and attention-line records (90% vs. 22%). The results suggest that multi-agent decomposition can improve the robustness and completeness of open-weight LLM extraction while preserving the privacy advantages of on-premise deployment. The study should be interpreted as a prototype evaluation on synthetic data rather than a production-readiness claim.