December 2025

Building PharmaGenesis AI: A Dual-AI Drug Discovery Platform

AI Drug Discovery Claude Gemini

PharmaGenesis AI started as an ambitious project to democratize drug discovery using AI. With support from AI Grants India, I was able to build a comprehensive platform that combines multiple AI models for pharmaceutical research.

🙏 Credits & Acknowledgments

AI Grants India - API access for Claude (aigrants.in, @aigrantsindia)
Google AI Studio - Gemini API access
Google Antigravity - Inspiration for the agentic AI coding experience
Claude Opus 4.7 - Primary AI for compound generation and validation

The 6 Implementation Phases

Phase 1: Export & 3D Visualization

Built PDF/CSV/JSON export utilities and integrated 3Dmol.js for interactive 3D molecular visualization. The viewer fetches structures from PubChem or generates from SMILES.

Phase 2: Favorites & ADMET Predictions

Implemented a favorites system with localStorage persistence and ADMET prediction engine for Absorption, Distribution, Metabolism, Excretion, and Toxicity analysis.

Phase 3: AI Follow-up & Comparison

Added an AI chat interface for asking questions about compounds, with quick actions like 'Refine', 'Explain Mechanism', and 'Suggest Alternatives'. Enhanced comparison view with multi-radar overlay.

Phase 4: Pipeline History & Synthesis Routes

Created auto-save functionality for all pipeline runs and a visual synthesis route diagram showing step-by-step chemical transformations.

Phase 5: Drug Interactions & Research Tools

Built Drug-Drug Interaction Checker with 8 common drug presets, Target Protein Information panel with links to UniProt/PDB/PubMed, and Literature Search for finding related research papers.

Phase 6: Clinical Trial Predictions & UX Polish

Added Clinical Trial Phase Predictor with success probability, timeline, and cost estimates. Implemented keyboard shortcuts for power users (J/K navigation, C for compare, ? for help).

Technical Stack

React + TypeScript for the frontend
Vercel for deployment with serverless API routes
Claude (Anthropic) for compound generation
Gemini (Google) for validation and analysis
3Dmol.js for molecular visualization
Recharts for data visualization

Key Learnings

The biggest challenge was handling CORS issues with direct API calls. I solved this by routing all AI requests through Vercel serverless functions, which also added security by keeping API keys server-side.

Try it at: pharmgenai.kprsnt.in