# I Finally *Saw* Backpropagation — By Building It From Scratch
I knew the maths but something felt off. So I built backprop from scratch using a Value class — and finally understood what PyTorch actually does.
Written by
Let us learn and build AI.
2nd year CSE (AI specialization) student building things I wish existed. I work at the intersection of LLMs, agentic AI, and full-stack development. I've built a multimodal agentic system with triple-LLM orchestration, a RAG-based study companion with spaced repetition and knowledge graph visualization, a full-stack habit tracker with XP and analytics, and I'm designing Aira — a modular AI OS connecting 8+ specialized agents through a shared orchestration layer. On this blog I write about what I actually build and learn — technical deep-dives into RAG pipelines and LangGraph workflows, honest build logs of projects in progress, and tutorials on the AI/ML concepts I'm figuring out as I go. I believe the best way to understand something is to build it, break it, and then write about why it broke. Stack: Python • LangChain • LangGraph • Flask • React • FAISS • MongoDB • SQLite • AWS
Responses
Markdown supported · ⌘↵ to post
No responses yet.