PinnedInTowards AIbyShehan Makani·11h agoPredicting Fluid Behavior Before the Lab: Physics-Based Screening From AI Data Centers to…Open models for PFAS-free immersion coolants and vacuum outgassing, and what they got right and wrong
InTowards AIbyShehan Makani·1h agoWhat Shap Revealed About Molecular Properties: Four Models, Four Different AnswersSolubility, boiling point, LogP and mutagenicity from SMILES, and why the top feature changes every time
Shehan Makani·11h agoI built a leakage-free gold price model. It still had three bugs.What we found when we re-checked our own walk-forward results, and why “tomorrow = today” is so hard to beat
Shehan Makani·Nov 10The Same Fluid Problem Sits Inside an AI Data Center and a SatelliteTwo studies on predicting fluid behavior before the lab, from immersion-cooled GPUs to spacecraft in hard vacuum
Shehan Makani·Jun 17Screening Molecules for a Type II Civilization: SpaceChem-AI Notebooks 06 + 07Forward screening at R² = 0.9956. Then inverse design, a graph neural network, and radiation qualification for Dyson swarm PDR. The full…
Shehan Makani·May 30When Chemistry Learns to Talk BackThe materials that self-heal, self-sense, and self-report — and what it means for specialty chemical supply chains
Shehan Makani·May 22We Built an AI That Screens Specialty Resins Before You Touch a Flask. Here’s What It Found.Five real blends. Five live API calls. Every result below came from production.
Shehan Makani·May 20I Built a Cheminformatics ML Series on Kaggle -Here’s What the Data Actually Told MeFour molecular properties. Four models. Four results that surprised me.
Shehan Makani·May 17Why Most Gold Price ML Models Are Lying to You - And How We Fixed ItA genuine next-day forecasting system: walk-forward validated, zero data leakage, real results across 121 folds.
Shehan Makani·May 2From Software to Steel: Building a Personal AI R&D Stack for Sustainable ChemistryFor a long time, most of the conversation around AI in industry has felt too abstract.