CausalSentinel

OpenSentinel Project Summary

Person & Context

Natalie Huang — UC Davis undergraduate with self-reported beginner Python skills, assigned a 2-week (~26 hour) internship building a drug-comparison agent.

Tasks & Deliverables

Primary deliverable: A working Gemini AI agent that accepts two drug names and produces comparison outputs.

Specific outputs required:

• Side-by-side comparison table of molecular properties and safety signals
• 3–5 sentence AI summary identifying property-safety patterns
• Minimum 2 example runs documented (e.g., aspirin vs. ibuprofen)
• README (screenshots of the app are sufficient; demo video optional/stretch)

Agent functionality: The system must autonomously call 2 database tools per submitted drug and assemble results into a structured table.

Tools & Skills Required

Database integrations (2 mandatory):

• PubChem PUG REST for molecular properties (MW, LogP, formula, etc. — pulled directly from the API response, no local RDKit computation needed)
• openFDA FAERS for adverse events

Technical stack: Python, Streamlit, google-generativeai, requests library

Programming paradigm: “Computation Engine — Streamlit form (paste drug names) → side-by-side comparison table”

Timeline & Structure

Week	Duration	Focus
1	~12h	Learn PubChem + FAERS APIs and Gemini basics; pair programming to build skeleton from starter template
2	~14h	Wire up Streamlit table + AI summary, test with 2 example runs, write README

Critical Success Factors

The plan explicitly depends on: (a) a shared starter template existing before week 1, and (b) Shucheng (PhD mentor) paired as technical guide through week 2.

Stretch Goals (optional, not required for MVP)

• ChEMBL REST for additional drug info
• openFDA Drugs@FDA for boxed warnings
• RDKit-based property computation

Out of Scope

BindingDB, LiverTox, toxicity classifiers, 50-drug calibration, withdrawn-drug validation, and 200+ drug pathways analysis—deliberately excluded to fit beginner timeline.

This site is open source. Improve this page.