Example Workflows
🌟 Featured: Personal Medical Assistant
A privacy-first, 100% local medical records helper that lets you understand your health data using AI and evidence-based medical knowledge.
✨ Highlights
- 🔒 Complete Privacy – All PDF processing in browser, no uploads, PHI stays on device
- 🤖 Multi-Hop Agentic RAG – AI orchestrates query generation and knowledge synthesis
- 📚 125,830 Medical Knowledge Chunks – From 18 authoritative textbooks (MedRAG dataset)
- ⚡ Two-Tier AI – Fast model for queries, capable model for responses
- 💬 Streaming Chat – Real-time responses with transparent reasoning
Built with Next.js and LlamaFarm, this example demonstrates how to build privacy-first healthcare applications that keep sensitive data completely local while delivering intelligent, evidence-based insights.
CLI-Based RAG Examples
The repository ships with interactive demos that highlight different retrieval scenarios. Each example lives under examples/<folder> and provides a configuration, sample data, and a script that uses the newest CLI commands (e.g., lf datasets create, lf chat).
| Folder | Use Case | Highlights |
|---|---|---|
large-complex-rag/ | Multi-megabyte Raleigh UDO ordinance PDF | Long-running ingestion, zoning-focused prompts, unique DB/dataset per run. |
many-small-file-rag/ | FDA correspondence packet | Several shorter PDFs, quick iteration, letter-specific queries. |
mixed-format-rag/ | Blend of PDF/Markdown/HTML/text/code | Hybrid retrieval, multiple parsers/extractors in one pipeline. |
quick-rag/ | Two short engineering notes | Rapid smoke test for the environment and CLI. |
How to Run an Example
# Build or install the CLI if needed
go build -o lf ./cli
# Run the interactive workflow (press Enter between steps).
# The script automatically scopes the CLI with `lf --cwd examples/<folder>`.
./examples/<folder>/run_example.sh
# Optional: point the script at a different directory that contains the lf binary
./examples/<folder>/run_example.sh /path/to/your/project
# Skip prompts if desired
NO_PAUSE=1 ./examples/<folder>/run_example.sh
Each script clones the relevant database entry, creates a unique dataset/database pair, uploads the sample documents, processes them, prints the CLI output verbatim, runs meaningful lf rag query and lf chat commands, and finishes with a baseline --no-rag comparison. Clean-up instructions are printed at the end of each script.
Manual Command Reference
Use these commands if you prefer to run the workflows yourself (replace <folder> with the example you want to explore):
lf --cwd examples/<folder> datasets create -s <strategy> -b <database> <dataset>
lf --cwd examples/<folder> datasets upload <dataset> examples/<folder>/files/*
lf --cwd examples/<folder> datasets process <dataset>
lf --cwd examples/<folder> rag query --database <database> --top-k 3 --include-metadata --include-score "Your question"
lf --cwd examples/<folder> chat --database <database> "Prompt needing citations"
lf --cwd examples/<folder> chat --no-rag "Same prompt without RAG"
lf --cwd examples/<folder> datasets delete <dataset>
rm -rf examples/<folder>/data/<database>
Refer to each example folder’s README for scenario-specific prompts, cleanup suggestions, and contextual background (e.g., why those documents were chosen and what use cases they simulate).