Molecular Data Chatbot

Designed and implemented the technical architecture for an Agentic AI system using Vertex AI ADK to enable bidirectional reasoning between molecular structure and adverse event risk. Developed a hybrid similarity model combining vector-space embeddings, Tanimoto fingerprint similarity, and regression logic to map SMILES inputs to AE risk and identify low-risk molecular analogs. Integrated large-scale biomedical datasets (ChEMBL, PubChem, FAERS) via API orchestration and web scraping, using RDKit, BigQuery, and Pandas to compute molecular descriptors and generate fingerprints. Deployed a cloud-native backend with Google Cloud Functions and GCS for real-time molecular inference, exposing a public HTTP API for seamless integration into full-stack pharmacovigilance applications.