A technical deep-dive into building Browser RAG, a 100% local, client-side retrieval-augmented generation engine running inside the browser using PGlite WASM, pgvector, WebGPU, and client-side LLMs.

An experiment in fine-tuning small, local language models (Qwen 1.5B) to reason in a telegraphic, compressed style, saving up to 74% of reasoning tokens and speed up inference without relying on verbose prompts.

A deep dive into exporting and optimizing AI4Bharat's IndicTrans2 models for local, offline translation in browser runtimes via ONNX, WebAssembly, and WebGPU.

A technical deep-dive into building WebVoice Studio, a fully local, zero-server voice AI workbench running entirely in the browser using WebGPU, WASM, Silero VAD, and custom model runner adapters.

An in-depth look at building a serverless-ready, zero-dependency Markdown-to-PDF converter in pure Python, deep-diving into AST mapping, Kroki caching, multi-pass layout engines, and typesetting safeguards.