Rakshit Sisodia enhanced reliability and onboarding for calcom/cal.com and microsoft/DeepSpeed by delivering targeted backend improvements. For calcom/cal.com, he implemented organization-scoped username validation, enabling seamless organization signups by resolving namespace conflicts. On microsoft/DeepSpeed, he strengthened checkpointing robustness through runtime validations, timeouts, and health checks, ensuring compatibility with the Nebula engine and reducing training downtime. He also addressed runtime stability by fixing numerical errors and crash scenarios in model training, such as replacing torch.sqrt with math.sqrt for dynamic batching and resolving NaN propagation in optimizers. His work leveraged Python, TypeScript, and deep learning frameworks, demonstrating strong backend engineering depth.