Lovable Candidate Knowledge Graph

Build and visualize a candidate knowledge graph backed by Neo4j + GDS with sentence-transformer embeddings and clean community detection. See similary between candidates and generated ideal profiles based on job opening descriptions. Built as an R&D project for candidate matching and recruitment at Lovable 6-7th September.

image.png

Prereqs

• Python 3.11+
• uv (pip install uv) - otherwise can use apt or package manager
• Docker with Docker Compose (recommended, for Neo4j + GDS)

Setup

uv venv
source .venv/bin/activate
uv pip install -r requirements.txt

Run scraping agent

uv run langflow run
python src/scrape/main.py

Run simple analysis to check that csv has been generated

python src/scrape/analysis.py

Note: sometimes the agent tweaks out and mismatches csv headers, drop rows that cause problems.

Start Neo4j (GDS)

python src/graph/setup_neo4j.py

Run analysis

python src/graph/test_graph.py

Outputs:

• candidate_knowledge_graph_gds.html (interactive graph)
• Data written to Neo4j (Browser at http://localhost:7474)

Tune communities (optional)

• Fewer, larger groups: lower edge threshold, raise top_k, or set target=3/4 for Leiden.

Config (env)

• NEO4J_URI, NEO4J_USER, NEO4J_PASSWORD (see .env created by setup)

TO DO

1. Add column type and classes "Human" or "Generated".
2. (OPTIONAL) Unfuck pipeline agent -> graph.
3. Script to find people similar to imaginary engineers.