Purvish Shah

Purvish Shah

Purvish Shah builds practical AI products and shares lessons from shipping real applications.

Chrome Extension: Political Bias Analyzer (Local LLAMA + OpenAI)

A Chrome extension that analyzes political bias in news articles and web content by extracting article content and providing a visual breakdown of left vs. right political leaning percentages. The ext

read more

Fine-Tuning GPT-2 on CFPB Dataset with ONNX and Gradio

What I LearnedWork with real-world datasets: Cleaning and preparing the CFPB dataset for model training. Dataset: [CFPB Consumer Finance Complaints](https://huggingface.co/datasets/CFPB

read more

From AI Anxiety to Shipping My First iOS App

Part 1 of a series documenting the journey of building an AI-powered personal growth app inspired by Hindu philosophy. Over the last few years, AI has changed how we work. Ideas that once needed a

read more

Multi-Tool AI Agent - Document Processing with LLMs, TTS & Image Generation

An intelligent, multi-tool AI agent that automatically processes documents using LLMs, TTS, and image generation, all orchestrated through a SQLite database and a Gradio web interface. What I Lear

read more

My Krishna Is Live: A New Way to Have a Conversation with Krishna

Today is a special milestone for me. My Krishna is live Getting My Krishna from an idea to the App Store was not easy. There were many things to figure out: building the application, setting up

read more

NASA APOD Explorer + AI Narrator (Gradio App with OpenAI)

A lightweight Gradio app that fetches NASA's Astronomy Picture of the Day (APOD) and generates a concise, human-friendly narration using OpenAI. It supports HD images, date selection, graceful handlin

read more

RAG for Tabular Datasets

What I Learned My goal for this project was to deeply understand Retrieval-Augmented Generation (RAG) using a tabular dataset, moving beyond unstructured text use cases. I worked with a r

read more

Sales Update AI Assistant (Running on HuggingFace - Space)

Sales leaders often receive long narrative updates from team members through email or chat. These updates contain useful information, but they are often inconsistent, difficult to compare week to

read more