How senthilmmurugesan/cnx-hackathon works
This repository is a Databricks drug-safety question-answering agent backed by adverse-event data. Its main workflow accepts a user question, uses a LangGraph tool-calling agent to retrieve drug and safety information, and returns a readable risk summary. Separate notebooks load source tables and prepare drug names for optional vector-based normalization. The excerpts show the agent structure and table loading, but do not expose the SQL function definitions or full tool wiring, so those links are documented rather than inferred from unseen code.
Main entry points: driver.py Drug safety agent · nb-create-vector.py Drug-name index · tables.sql Safety tables · nb-create-input-tables.py Input table loader
Components 10
Agent runtime
- Drug safety agent
driver.py - LangGraph workflow
driver.py - Serving model
- Retrieval tools
driver.py
Safety data
- Safety SQL functions
functions.sql - Safety tables
tables.sql - Drug-name index
nb-create-vector.py
Data preparation
- Input table loader
nb-create-input-tables.py - Tool notebook
nb-create-tools.py
Other components
- Questioner
Connections 9
- Questioner → Drug safety agent: asks
driver.py - Drug safety agent → LangGraph workflow: runs
driver.py - LangGraph workflow → Serving model: invokes
driver.py - LangGraph workflow → Retrieval tools: dispatches
driver.py - Retrieval tools → Safety SQL functions: queries
driver.py - Retrieval tools → Drug-name index: normalizes
driver.py - Drug safety agent → Questioner: answers
driver.py - Input table loader → Safety tables: writes
nb-create-input-tables.py - Drug-name index → Safety tables: reads
nb-create-vector.py