Rushabh Doshi is director of sales and marketing at Machinecraft, his family’s third-generation Indian thermoforming-machinery manufacturer, and the creator of Ira, an AI system that turns institutional knowledge into usable operational intelligence. His central insight is that an industrial company’s most valuable asset may be its accumulated customer relationships, technical specifications, quotations, and commercial judgment—and that existing AI models can use this knowledge without expensive proprietary training.
Doshi studied automation and robotics at King’s College London and joined Machinecraft’s commercial leadership in 2013. He also completed management training at RWTH Aachen University, founded AESOP Design, and co-founded Kodex OS. As the India representative for German manufacturing-technology company FRIMO, he has helped connect thermoforming with complementary production techniques, including polyurethane back-foaming.
Machinecraft’s equipment serves industries ranging from automotive and medical manufacturing to packaging and agriculture. That variety made institutional memory unusually valuable: customer histories, machine modifications, quotations, and business relationships were distributed across documents, email threads, and individual employees. Doshi built Ira to recover that context and apply it to sales and customer operations.
How Ira works
- Organizational memory without model training: Ira extracts knowledge from existing business records and combines vector search, relationship graphs, and customer data instead of training or fine-tuning a proprietary model.
- Specialist agents with human oversight: Athena coordinates narrowly scoped agents responsible for sales, pricing, machine specifications, fact-checking, and preserving human corrections. Their work supports account research, quotations, lead qualification, and draft outreach; people approve outgoing communications.
- Nightly memory consolidation: The system revisits recent activity, retains useful facts, identifies contradictions, removes outdated information, and develops reusable capabilities. Human corrections take precedence when records disagree.
- Forkable organizational infrastructure: Doshi extended the architecture into BrainOS, a reusable foundation for organization-specific agents, persistent memory, and behavioral rules emphasizing verification, uncertainty, accountability, and clearly defined responsibilities.
Through his AI-agent practice and AI Engineer World’s Fair presentation, Doshi advances a practical approach to enterprise AI: make a company’s accumulated knowledge persistent, intelligible, and available when operational decisions demand it.
