Sample data

Three sample documents so anyone can run RagFlowPlus and judge its performance without supplying their own files. Two are small hand written business documents, one is a real public SEC filing, so you can see it work on both clean text and a dense real world document.

File	Type	Use case
acme_employee_handbook.txt	HR policy	Internal policy lookup
nimbus_product_faq.txt	Product FAQ	Support and pricing questions
nvidia_10k_excerpt.txt	Real SEC 10-K excerpt	Finance questions over a real document

Load them

Fully local and keyless with Ollama:

ollama serve &
ollama pull nomic-embed-text
ollama pull llama3.2:3b
EMBEDDING_PROVIDER=ollama python scripts/load_sample_data.py

Then start the UI and ask a question.

Questions to try, with the expected answer

Handbook:

• How many vacation days do new employees get? (15, rising to 25 after five years)
• What is the remote work policy? (up to three days per week)

Product FAQ:

• What does the Nimbus Pro plan cost? (49 dollars per month billed annually)

NVIDIA 10-K:

• What are the building blocks of NVIDIA's platform? (GPUs, CPUs, CUDA, networking)

Honesty check (not in any document):

• What is Acme's parental leave policy? (it should say the documents do not cover this)

What RagFlowPlus adds over the naive baseline

RagFlowPlus is the 2023 Advanced rung. Unlike the naive RagFlow, it fuses dense and BM25 keyword search with Reciprocal Rank Fusion and reranks the result with a cross encoder, so answers to harder questions are better grounded.