CatalystMind AI —AI for Bharat 2026 — Theme 4 (Prototype Round)

CatalystMind AI — Molecular Discovery for Catalysis & Synthetic Biology (GPS Renewables) PanIIT AI for Bharat 2026 — Theme 4 (Prototype Round) Discovery for sustainable fuels is slow, expensive, and siloed across two parallel R&D tracks — chemical catalysis and synthetic biology. Different teams, different tools, different data models. Every novel candidate costs lab time and money, every failed experiment is treated as wasted, and there's no shared retraining loop. CatalystMind AI is built for GPS Renewables — one platform, two reaction families, three end-to-end capabilities. Generate, Narrate, Learn. What's in this 5-minute walkthrough: • Dashboard — AI Research Briefing + activity × selectivity scatter (blue = DB known, dark = AI-generated novel) clustering in the high-activity quadrant • Reaction workbench — pick CO₂ → Methanol (GPS Renewables E2J pilot) - Pathway tab: AI mechanistic free-energy narration + SVG diagram of 5 idealised states (reactant → intermediates → TS → product) - Candidates tab: ranked by activity / selectivity / stability / yield • Candidate detail — pick a novel composition - 3Dmol.js viewer renders the SMILES structure interactively - AI Candidate Rationale (Azure GPT-4.1) explains why this molecule, what each prediction means, what risks an SME should probe • Models — Retrain endpoint is a real Jaccard-similarity algorithm, not a flat accuracy bump. Walks every logged experiment, computes (measured − predicted) deltas, propagates corrections to peers (factor 0.3, min similarity 0.34). Reaction-type is treated as a feature, not a model selector — same pipeline, same retrain loop, both directions. Chapters 0:00 Intro + Problem 0:30 Solution 0:55 Research dashboard 1:35 Reaction workbench — AI pathway 2:30 Candidates — rationale + 3D 3:20 Retrain feedback loop 3:50 Grounded AI + No Shortcuts 4:15 Demo data + Risks 4:35 Roadmap + Why We Win Brief non-negotiables met: synthetic data only · physics-based mock predictor (DFT-swappable in production) · grounded LLM narration only · explainable retrain · SME-in-the-loop. Stack: Next.js 16 + Prisma + SQLite, Azure OpenAI GPT-4.1 with deterministic mock fallback, 3Dmol.js for 3D structure, Tremor charts. 🔗 Code: github.com/sridhar7601/catalyst-mind-ai 👥 Team: Sridhar Suresh + Sruthi Krishnakumar 🏆 PanIIT AI for Bharat Hackathon 2026 — Theme 4 (GPS Renewables Molecular Discovery) #PanIIT #AIForBharat #GPSRenewables #Catalysis #SyntheticBiology #CatalystMindAI