FatimaGlow HR Helpdesk

This project adapts the RAG/hr-helpdesk/experiments reference into a runnable HR helpdesk for the policy files in docs/.

What it includes

• Markdown-aware chunking for the numbered policy sections in your docs
• PostgreSQL + PGVector for vector indexing
• Google Gemini embeddings and chat generation
• Streamlit chat UI with source citations

Setup

1. Create a virtual environment and install dependencies.
2. Ensure .env contains GOOGLE_API_KEY.
3. Start PGVector locally:

docker compose up -d

4. Wait for the pgvector container to become healthy.
5. Build the index:

python -m hr_helpdesk.indexing

6. Run the app:

streamlit run hr_helpdesk/main.py

Environment variables

• GOOGLE_API_KEY required
• EMBEDDING_MODEL optional, defaults to gemini-embedding-001
• CHAT_MODEL optional, defaults to gemini-2.5-flash
• HR_DOCS_DIR optional, defaults to docs
• HR_COLLECTION_NAME optional, defaults to fatimaglow_hr_helpdesk
• PGVECTOR_CONNECTION required for PGVector, for example postgresql+psycopg://postgres:postgres@localhost:5432/fatimaglow_hr
• EMBEDDING_DIMENSION optional if you want a fixed embedding width for PGVector

Local PGVector

This repo includes compose.yaml and 01-enable-vector.sql for a local development database.

Default connection:

postgresql+psycopg://postgres:postgres@localhost:5432/fatimaglow_hr

Useful commands:

docker compose up -d
docker compose ps
docker compose down