SK-404

Doc no. SK-404DatasheetRev 2026.11Sheet 1 of 4

Shamit

S-k404 ECE undergrad

High tech, low overhead. I build local-first software, hardware and self-hosted systems that run on machines I own.

ProjectsGitHub

Doc no.
SK-404
Rev
2026.11
Sheet
1 of 4
Base
Bengaluru, IN

Selected work

  • 12 V / 100 W through-hole PCB heater driver that latches off on over-current and over-temperature.

    100 W at 12 V

    • KiCad
    • Through-hole
    • ngspice

    Prototype

  • Printed 10-inch, 9U mini rack built around a Raspberry Pi 5, with print files, STEP and Onshape CAD and a build checklist.

    9U 10-inch rack

    • 3D printing
    • CAD
    • Raspberry Pi 5

    Prototype

  • Turns textbooks and technical PDFs into chaptered .m4b audiobooks — entirely on-device, on the Apple Silicon GPU.

    ~16× real-time synthesis

    • Python
    • PyTorch
    • PyMuPDF
    • FastAPI

    Shipped

  • Glue and tools for a local, OpenAI-compatible speech engine in Docker, built for Odysseus: Kokoro-82M speech, Faster-Whisper transcription and document-to-audio scripts.

    Built for Odysseus; Runs on Speaches

    • Docker
    • Kokoro-82M
    • Faster-Whisper
    • Python

    Active

  • Exports a Notion workspace into an Obsidian vault with a true folder hierarchy, resolved relations and native Obsidian Bases.

    2.8 req/s, paced under the limit

    • Python
    • Notion API
    • httpx
    • PyYAML

    Shipped

  • Batch downloader for videos, image posts and files: a CLI, a browser GUI, a Mac app and a stream-catching extension, built on yt-dlp, gallery-dl and aria2c.

    • yt-dlp
    • gallery-dl
    • aria2c
    • FFmpeg

    Active

All 22 projects

About

I'm an electronics & communication engineering undergrad whose work sits on both sides of the stack: HDL, FPGA and PCB toolchains on one side, Python services, containers and GPU inference on the other.

Most of what I build starts as a tool I wanted for myself, then gets the boring parts done properly: dry-runs, rate limits, atomic writes, a self-diagnostic, tests.

Features

  • Runs ML inference on local silicon: no API keys, no per-token bill
  • Comfortable in MATLAB, Vivado and Quartus as well as the shell
  • CLIs that fail safe: dry-run first, never half-write a file
  • Self-hosts in a homelab, containerised

What is next

No grand plan, just a build queue. What I want to spend the next stretch making and learning:

  1. Ongoing

    Edge & local AI

    Keep pushing models onto hardware I own: faster, smaller, and eventually onto low-power devices.

  2. Ongoing

    Automation & self-hosting

    Grow the homelab and automate the everyday: more services run myself, fewer manual steps.

  3. Ongoing

    Open-source tools

    Keep shipping CLI tools like the ones on the projects sheet: useful, careful, tested.

  4. Next

    Hardware & FPGA

    Go deeper into digital design on FPGAs with Vivado and Quartus, and lay out my own boards in KiCad.

  5. Planned

    Physical builds

    A cyberdeck, and more 3D-printed parts and enclosures. Things you can hold, not just run.

On the bench

  1. Mac Monitor

    Mac temperatures and battery health on a dashboard any device on the LAN can open.

    80% verifying

  2. Home NAS

    Network storage for the homelab.

    25% planning

  3. Mini HomeLab

    A small mini-PC compute box for self-hosted services.

    25% planning

All 29 builds on the Bench

Reference picks

Other people's builds I keep going back to, and what each one is good for.

  • Tiny Cyberdeck in an Altoids Tin

    The best-documented power design of the tin decks: charger, boost and fuel gauge on one board, with BOMs, PCBs and the enclosure in the repo.

  • Pocket Cyberdeck in a Barkleys Mint Tin

    An ESP32 instead of a Pi, with a GUI, terminal, file browser and emulator in firmware. Check the FPC connector orientation before ordering.

  • A Curiously Minty Cyberdeck

    The keyboard firmware (KMK on CircuitPython) and key mapping are public in the repo. Shows how to stack parts and desolder connectors to make them fit.

Connect

The fastest way to reach me or see what I'm building is GitHub.

GitHub