TALAAN: A Framework for Offline Multi-Tier Retrieval-Augmented Generation in Philippine Local Government Document Intelligence

Abstract—Philippine Local Government Units (LGUs) accumulate thousands of legislative documents—resolutions and ordinances—that remain largely inaccessible for evidence-based policy analysis due to fragmented storage, inconsistent naming conventions, OCR degradation, and the absence of semantic search infrastructure. This paper presents TALAAN (Tagged Archival Lookup for Analysis and Accountability Network), a framework for offline Retrieval-Augmented Generation (RAG) designed for LGU document intelligence under resource-constrained, privacy-sensitive deployment conditions. TALAAN does not attempt to replace the full document and records lifecycle; it operates as a retrieval and synthesis layer on top of existing document holdings. The framework introduces a seven-stage processing pipeline covering PDF ingest, multi-layer OCR correction with Filipino diacritics restoration, three-method metadata extraction, and a dual-table tagging architecture that enables systematic comparison of keyword-baseline versus LLM-generated tags. Retrieval is handled by a priority-enforced three-tier architecture combining deterministic exact-match (T1), structured SQL tag and metadata filtering (T2), and semantic vector search via ChromaDB (T3) at a calibrated cosine threshold of 0.75. The system operates fully offline on commodity CPU hardware using Ollama with gemma4 and incorporates privacy controls designed to support compliance with the Philippine Data Privacy Act of 2012. While existing platforms such as NotebookLM and Gemini offer document AI capabilities for LGUs already using Google Drive, these require document transmission to external servers—requiring institutional assessment of privacy and data-processing obligations under RA 10173 and operational dependency on reliable internet connectivity. TALAAN's locally-sovereign design addresses both constraints. A working implementation has been deployed on a corpus of 1,867 Sangguniang Bayan resolutions and ordinances from the Municipality of Midsayap, Cotabato (2021–2025).