Palash Debnath

Agentic Software Designer and Software Engineer

hi@palash.dev · palash.dev · github.com/debpalash · x.com/idebpalash

Agartala, Tripura, India — open to remote · IST (UTC+5:30)

35k+ GitHub stars · 121k+ release downloads · 6 shipped products · 10 articles

Open to: Senior Software Engineer; AI / ML Engineer; Systems Engineer (Rust / Zig); Full-Stack Product Engineer

Summary

Founder of YUPCHA, based in Agartala, India. I design AI agents and build open-source, local-first desktop apps and developer tools: 6 shipped products with 35k+ GitHub stars and 121k+ release downloads. I own the full loop: systems work in Rust, Zig, and CUDA; desktop apps; web on TypeScript and Cloudflare Workers; releases and support. Flagship: VoiceStudio.sh, an open-source local AI voice studio (35k+ stars).

Skills

Languages: TypeScript, Rust, Zig, Python, Go, SQL
AI / audio: TTS + ASR pipelines (16 TTS / 11 ASR engines), Whisper, CUDA, MLX, llama.cpp, MCP servers, agent memory (sqlite-vec)
Systems: Tauri desktop apps, GPU offload (CUDA / MPS / ROCm), Docker, static binaries, UEFI / boot media
Web: SolidJS, Cloudflare Workers (streaming SSR), Vite, edge caching, SEO / structured data

Experience

Founder and Independent Open-Source Builder

Present

YUPCHA — local-first, privacy-respecting software

Ship and maintain 6 open-source products and 19 smaller public repos; 35k+ GitHub stars and 121k+ stable-release downloads combined.
Run the full product loop: architecture, implementation, docs, releases, and community support.
Write about engineering and AI at palash.dev (10 articles).

Selected open-source work

VoiceStudio.sh - open-source local AI voice studio

35k stars - 120k downloads - v0.5.6, Sep 2026

Make voices. Tell stories. Keep the files.

Desktop studio for voice cloning, voice design, generation, dubbing, dictation, and audiobooks; runs locally, no subscription.
Engine layer: 16 TTS + 11 ASR engines across 646 languages on CUDA, Apple MLX, ROCm, or CPU; OpenAI-compatible local API and MCP server.

Stack: Python, Tauri, CUDA, Docker, MLX, Whisper. https://github.com/debpalash/VoiceStudio

Bootable - cross-platform boot media writer (Rust)

7 stars - 25 downloads - v0.1.1, Aug 2026

Cross-platform boot media writer.

Native desktop app and mouse-enabled TUI; removable-media-only guards, hash and byte verification, cancellable writes; Linux, Windows, macOS.

Stack: Rust, GPUI, Ratatui, UEFI. https://github.com/debpalash/bootable

memXT - local-first memory for AI coding agents

33 stars - 67 downloads - v0.4.0, Jul 2026

Local-first long-term memory for AI coding agents.

Single static Zig binary: MCP tools and session hooks, local embeddings (llama.cpp), sqlite-vec search; ~10 ms wake-up, sub-ms warm search.

Stack: Zig, LLM, MCP, AI Memory. https://github.com/Yupcha/memXT

Opal - AI media player (Zig)

64 stars - 1.1k downloads - v0.8.5, Sep 2026

Play everything — the evolved media player.

One native window for files, sites, servers, and feeds; natural-language request-to-playback; plugin endpoints evolve without growing the core.

Stack: Zig, AI, Desktop. https://github.com/debpalash/Opal

OpenGTM - self-hosted lead sourcing and enrichment

20 stars

Self-hosted Clay alternative.

Lead sourcing, enrichment waterfalls, AI research, and activation in one self-hosted workspace; bring-your-own keys with cost estimates before each run.

Stack: Python, FastAPI, React, PostgreSQL, Docker, MCP. https://github.com/debpalash/OpenGTM

Friday - local-first voice assistant for Linux

7 stars - alpha

Local-first voice assistant for Linux.

Voice and text assistant on local models (Qwen, Parakeet, Piper) with SQLite memory and bounded tool families; no required hosted calls.

Stack: Python, FastAPI, Qwen, Parakeet, Piper, SQLite. https://github.com/debpalash/friday

Other public repos: omnivoice-gallery, FullStack-AI-Engg-45days, agentwork, omarchy-bootable, omarchy-site-submission, hoglet — see the Lab.

Writing

10 articles at palash.dev/blog.