Project: ACUÑAPEDIA - Full-Stack Web Development | Pool Deza
Details, tech stack, key features, and impact of ACUÑAPEDIA
Intelligent RAG System with Auto-Sync, Hybrid Search, and Intent Analysis for the Political Sector.
ACUÑAPEDIA is a next-generation Retrieval-Augmented Generation (RAG) system, designed to process and answer queries about government plans intelligently and in real time.
This project was a high-level technical collaboration: Irvin Acuña (YairAcuna3) developed the base architecture and containerization, while I (iPool23) implemented production optimizations, advanced features, and ongoing system development.
The system integrates data from multiple sources (PDFs, press articles, and MySQL), generating a knowledge base of over 22,500 embeddings with hybrid search and an intelligent Query Analyzer that automatically detects user intent.
- Lead AuthorIrvin Acuña
- ContributorPool Deza
- Year2025-2026
- Embeddings22,564 Indexed
- Latency1.2s (Streaming)
Optimization of searches combining vector embeddings with BM25 and dynamic reranking for maximum precision.
Automatic detection of user intent and category to adjust context and search top_k.
Ingestion pipeline that synchronizes Press and Works records directly from relational databases.
Real-time response implementation to improve UX, reducing perceived latency from 8s to 1.2s.
Implemented adaptive chunking and asynchronous batch processing to prevent out-of-memory (OOM) errors on large files.
Trained the QueryAnalyzer to detect automatic context and dynamically adjusted BM25 weights for real data.
Configured a batch size of 128 and hardware optimizations for production deployment on EC2 without exclusive dependency on GPUs.