cortex / README.md
cortex-deploy
Deploy 5c5fcc1 from claude/neo4j-vs-falkordb
ac156d2
|
Raw History Blame Contribute Delete
2.37 kB

A newer version of the Gradio SDK is available: 6.32.0

Upgrade
metadata
title: Cortex
emoji: 🧠
colorFrom: blue
colorTo: yellow
sdk: gradio
sdk_version: 6.29.0
python_version: '3.12'
app_file: app.py
pinned: false
license: mit
short_description: Why FalkorDB for AI engagements, and more

Cortex: Why FalkorDB for AI engagements, and more

GraphRAG demo on Cortex Bank, a synthetic 100-year-old bank. 73 bank PDFs are ingested with LangChain into FalkorDB. The same questions are answered by plain vector RAG and by GraphRAG, with role-based access control, tokenomics, one-click scenarios (regulatory impact, fraud rings, risk roll-up, lineage, performance) and private upload sandboxes. Neo4j vs FalkorDB: a generic AML graph loaded identically into both engines, with role-based queries timed live and answers compared. The Engineer guide tab explains how to reproduce it inside your own network. All data is synthetic.

Run locally

cd cortex
python3.12 -m venv .venv && . .venv/bin/activate
pip install -r requirements.txt gradio==6.29.0
python app.py            # starts embedded FalkorDB, builds both graphs, serves http://localhost:7860

Optional environment variables:

Variable Purpose
GOOGLE_API_KEY Shared Gemini key (free tier) for the demo
CORTEX_ACCESS_PIN PIN testers enter to use the shared key
CORTEX_GEMINI_MODEL Gemini model name (default: newest Flash model available to the key, discovered automatically)
CORTEX_OPS_SCALE Operational graph size (1.0 = ~1M nodes; auto-sized from RAM when unset)
FALKOR_HOST / FALKOR_PORT Use an existing FalkorDB server instead of the embedded one

Layout

Path What
content/bank.py The bank's knowledge base (single source of truth and answer key)
generator/ PDF renderer and operational data generator (seeded, with fraud rings)
pdfs/ The 73 generated PDFs
ingest/ LangChain pipeline: PDFs to chunks, embeddings and graph (pattern + LLM extraction)
core/ Retrieval (vector RAG vs GraphRAG, role filter), LLM gateway with tokenomics, scenarios
app/ FastAPI server, limits, PINs, uploads, static UI
eval/ Extraction accuracy check and question set