📌 Project Overview

ERES is a local-first, high-performance RAG (Retrieval-Augmented Generation) pipeline designed to transform complex Emergency Operations Plans (EOPs) into actionable intelligence for field responders. The system ingests lengthy, multi-format planning documents and provides instant, grounded answers during active incidents.

The platform bridges the gap between static, PDF-based plans and the real-time information needs of emergency responders operating under high-stress conditions.

🎯 Problem Statement

In Emergency Management, Emergency Operations Plans (EOPs) are critical but cumbersome. Often spanning hundreds of pages, these PDFs are difficult to navigate under the high-stress conditions of an active incident.

Current EOP Limitations:

  • Accessibility: Critical information buried in hundreds of pages of documentation
  • Search Friction: No effective way to quickly locate specific procedures or contact information
  • Visual Data Loss: Maps, organizational charts, and diagrams ignored by standard text extraction
  • Time Pressure: Responders need immediate answers when seconds matter

ERES addresses these challenges by providing intelligent document retrieval that preserves both textual and visual information while maintaining complete source accountability.

✨ Key Capabilities

  • Visual Preservation: Go-based extractor archives maps, ICS charts, and diagrams for situational awareness
  • Layout-Aware Ingest: Sophisticated partitioning understands document hierarchy (headers, sections, tables)
  • Fully Local & Private: Designed for air-gapped or sensitive government environments; no data leaves the local machine
  • Source Accountability: Every answer includes direct page citations from the original EOP to prevent AI hallucinations
  • Dual-Engine Architecture: Separate processing for text and visual data ensures nothing is lost
  • High-Performance Retrieval: Optimized vector search for sub-second response times

🏗 High-Level Architecture

The system utilizes a unique dual-engine architecture to ensure both text and visual data are preserved and utilized:

ERES Architecture

Architectural Principles:

  • Extraction Layer: Go-based high-speed PDF parsing with visual asset preservation
  • Intelligence Layer: Python-based RAG pipeline with LangChain orchestration
  • Retrieval Layer: Vector database for semantic search and embedding storage
  • Inference Layer: Local LLM for grounded, citation-backed responses

This design ensures complete information preservation while maintaining privacy and performance requirements for government use cases.

🛠 Tools & Technologies

Layer Technology Purpose
Language (Extraction) Go High-speed PDF parsing and binary portability
Language (AI) Python 3.12 LangChain orchestration and data science libraries
AI Framework LangChain RAG pipeline orchestration and prompt management
AI Model Llama 3.2 (3B) Local LLM served via Ollama for low-latency inference
Vector Database ChromaDB High-performance embedding storage and retrieval
Model Serving Ollama Local model deployment and inference optimization
Development WSL2 (Ubuntu) Linux-standard development on Windows hardware
DevOps Docker Containerization for reproducible deployments

🚀 Future Roadmap: Cloud Scale

Currently transitioning the architecture for event-driven cloud processing:

  • Cloud Infrastructure: AWS Lambda for serverless, event-driven document processing
  • Storage: S3 bucket triggers for automatic intelligence pipeline activation
  • Scalability: Automatic scaling for multi-document and multi-user scenarios
  • Monitoring: CloudWatch integration for pipeline observability and performance tracking