▶ Videos →

📰 Dev.to · Tushar Jaju

5 articles · Updated every 3 hours · View all reads

All Articles 131,075 Blog Posts 136,076 Tech Tutorials 33,950 Research Papers 25,448 News 18,608 ⚡ AI Lessons

My library told every Hindi user their model was broken

Dev.to · Tushar Jaju 3d ago

My library told every Hindi user their model was broken

llmclean 0.4.0 — how a Latin-only assumption flagged 100% of Hindi and Chinese output as truncated, why my unit tests could never have caught it, and what I bui

Whisper transcribes spoken Hindi numbers as words, not digits. So I made a library to convert them back.

Dev.to · Tushar Jaju ⚡ AI Lesson 2w ago

Whisper transcribes spoken Hindi numbers as words, not digits. So I made a library to convert them back.

Why I extracted hindi-normalize from a health-worker voice pipeline — a zero-dependency Python library for turning messy Hindi ASR/OCR text into clean, parseabl

I almost added an em-dash remover to my LLM library. Then I tested whether local models even produce em-dashes.

Dev.to · Tushar Jaju 🧠 Large Language Models ⚡ AI Lesson 1mo ago

I almost added an em-dash remover to my LLM library. Then I tested whether local models even produce em-dashes.

How a five-model local sweep reshaped llmclean 0.3.0 — and three things about LLM output that turned out to be the opposite of what I assumed.

Building Sakhi: Hindi Voice-to-Form for India's ASHA Workers, Solo in Six Weeks

Dev.to · Tushar Jaju 💻 AI-Assisted Coding ⚡ AI Lesson 2mo ago

Building Sakhi: Hindi Voice-to-Form for India's ASHA Workers, Solo in Six Weeks

TL;DR — Six-week solo build of a Hindi voice-to-form pipeline for India's ~1 million community...

I kept rewriting the same regex passes against LLM output. So I made a library.

Dev.to · Tushar Jaju 🧠 Large Language Models ⚡ AI Lesson 2mo ago

I kept rewriting the same regex passes against LLM output. So I made a library.

Why I built llmclean — a tiny Python library for cleaning LLM output, and what production traffic across three of my projects taught me to fix in v0.2.0.