Paritosh Baghel
Software craft, system design, reliability.
Architect and software engineer writing about what actually works — drawn from a decade of building products across startups, consulting, and product companies.
Featured essay
Why Neural Networks Learn: Loss Functions and the Taylor ApproximationFrom first principles: what it means for numbers to 'learn', why we measure wrongness the way we do, and how a 300-year-old calculus trick powers all of modern optimization. With the math derived and the code written out.
Recent writing
- The Kernel Trick, or: Geometry You Never Have to Visit
- Manifolds: The Low-Dimensional Lie Inside Your Embeddings
- Why Everything Wants to Be Gaussian
- The Shape of Infinite Dimensions
- Leases Are the Missing Primitive for Multi-Agent Coordination
- Reading Code That a Model Wrote: A Field Guide
- A Short History of Thinking in Many Dimensions