manavv

toy-qwen3. A minimal character-level implementation of the Qwen3 architecture in PyTorch, trained to generate Shakespearean text.

first-RAG. A RAG application built from the official LangChain tutorial. Features intelligent query analysis, section-based retrieval, and interactive CLI for document Q&A.

YodaLM. A fine-tuned language model that speaks like Master Yoda, built with Unsloth + QLoRA on Qwen3-0.6B.

mnav. portfolio website

toy-qwen3-MoE. A minimal character-level implementation of the Qwen3-MoE architecture in PyTorch, trained to generate Shakespearean text. Built from scratch for learning purposes.