Himanshu Jha / Full-stack AI developer

I build AI
people can use.

From local models to full-stack products, I turn ideas into working software with clear controls and inspectable results.

Himanshu's illustrated GitHub avatar

Selected work.

Local AI, practical tools, and a little theatre.

FrameKind running in Safari, detecting a person, bicycle and potted plant beside an editable alt-text draft

FrameKind

On-device AI

Useful alt text. Your image stays yours.

Browser-based object detection turns images into editable alt-text drafts, with on-device benchmarks to check the trade-offs.

ReactTransformers.jsWebGPU
Project details

YOLOS Tiny runs in a dedicated Web Worker. The app compares inference configurations against full-precision detections before choosing a default. Drafts describe detected objects and their positions; they still need human review.

The real OmniDev macOS app with its local Gemma chat and developer tool sidebar

OmniDev

Developer tools

A local AI workspace for your Mac.

A native macOS assistant that reads files, cites its sources, edits code and runs tests using local models.

SwiftUIPythonOllama
Project details

A SwiftUI interface connects to a local Python engine for retrieval, OCR and tool execution. Ollama supports offline use; a separate Gemini integration is optional. The source and unsigned macOS builds are available on GitHub.

The Forgetting Stage web lobby with its premise and instructions to direct a play and pin one truth

The Forgetting Stage

Creative AI

Direct a cast with a fading memory.

An AI improv game where you preserve one fact while the cast forgets and rewrites the rest.

ExpoReact NativeRevenueCat
Project details

The cast shares a bounded memory, and the player chooses a single fact to preserve. The final drift report compares the original premise with what the actors now believe. Built as a successor to Thousand-Token Theater.

A little about me

Curious about models.
Particular about
the product.

I’m Himanshu, a full-stack AI developer. I build agent systems, developer tools, and applications that run AI on your own device.

My projects span Python backends, React interfaces, and native macOS apps. I care about what happens after the model answers: how you check it, correct it, and decide what happens next.

Product & interfaces

TypeScript, React, Next.js, SwiftUI

Models & systems

Python, FastAPI, MCP, local inference

Explore the repositories

AMD DevMaster Excellent Award

Team himanshu748 was named among the top eight teams in AMD’s 2026 DevMaster Hackathon.

Read AMD’s recap

Writing about the work.

Build notes, technical decisions, and what I learned.

More on DEV

Have something
worth building?

Tell me what you’re working on.

Contact