What It Does in Plain English
This system acts like a specialized search engine for your own PDF documents. It reads, understands, and answers your questions with precise, source-verified information.
Reads PDFs
Ingests and comprehends the content of your documents.
Remembers
Stores information in an intelligent vector database.
Answers
Finds relevant context to answer your specific questions.
Shows Sources
Provides exact page and location for every piece of data.
Two Powerful Ways to Use It
Whether you need instant answers or a comprehensive report, the system adapts to your workflow.
Interactive Mode (Chat)
A real-time, conversational interface for exploring your documents. Ask questions and get immediate, sourced answers.
- ✓ Perfect for learning and discovery
- ✓ Immediate feedback loop
- ✓ Shows page numbers and locations
Batch Mode (Report)
Automatically processes a list of questions and saves all answers to a structured file for later analysis.
- ✓ Ideal for comprehensive analysis
- ✓ Run large jobs unattended
- ✓ Generates shareable reports
Understanding System Capacity
While there's no hard limit on the number of PDFs, performance depends on document size and available hardware resources.
Estimated Processing Time per PDF
GPU Memory Limit
5-10
Large PDFs (100+ pages)
System Memory Limit
10-20
Medium PDFs (10-50 pages)
Test Benchmark
1,737
Nodes from 5 PDFs
Quick Start on the HPC
Follow these four steps to get the system up and running on the Dartmouth HPC cluster.
Step 1: Upload Files
Transfer scripts and PDF documents to your project directory on the HPC using `scp`.
Step 2: Setup Environment
Run the setup script to create the Conda environment and install all necessary packages.
bash setup_env.shStep 3: Test the System
Launch an interactive session or submit a batch job to process your documents and queries.
srun --partition=gpuq ... bash
sbatch run_batch.slurmStep 4: Verify Results
Check the output logs and results directory for generated answers, sources, and performance statistics.
System File Structure
A well-organized project with core logic, documentation, and versioned archives for maintainability.
Core System
- interactive_rag.py
- batch_rag.py
- setup_env.sh
- run_interactive.sh
- run_batch.slurm
Documentation
- README_RAG_SYSTEM.md
- RAG_SOLUTION_WORKING.md
- RAG_DEPLOYMENT_GUIDE.md
- UPLOAD_AND_RUN.md
Archive
- rag_versions/
- VERSION_HISTORY.md
- debug_scripts/