Skip to content
View Pratham2994's full-sized avatar

Highlights

  • Pro

Block or report Pratham2994

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please donโ€™t include any personal information such as legal names or email addresses. This note will only be visible to you.
Report abuse

Contact GitHub support about this userโ€™s behavior. Learn more about reporting abuse.

Report abuse
Pratham2994/README.md

About

  • ๐Ÿฆ Software Engineer at Barclays, mainly Python and TypeScript/JavaScript (React, Node, Express), with C++ when performance is the actual constraint
  • ๐Ÿค– Running local models and agent harnesses, and measuring what breaks
  • ๐Ÿโšฝ๐ŸŽ๏ธ๐Ÿธ๐ŸŽพ๐ŸŽฎ Off the keyboard: sports, all of it, cricket, football, F1, badminton, tennis, pickleball, squash, and gaming
  • ๐ŸŒ Everything I have built, with a demo on each: prathampanchal.dev

Tech Stack


Projects

OmniCompiler

Run, step-debug and translate code across Python, JavaScript, Java, C++ and Go from one interface. Debugging is driven by the real per-language debuggers (gdb, jdb, pdb, Delve) behind a single event protocol, not a simulated stepper. Every run gets its own Docker container. Research paper under review.

Scrub

Local-only ffmpeg GUI, no server, no account, nothing leaves your machine. Give it a Discord, WhatsApp or email size limit and it two-pass encodes to land under it, instead of leaving you to guess a bitrate. The command it builds is shown on screen before it runs, and it's editable, so what you see is exactly what executes.

LocaLLM

Which local model is worth running on one 12 GB card? A test harness around llama.cpp: 34 tasks of my own, 41 configurations, 4,101 runs, each answer checked by running tests or comparing SQL rows. It ranks models by correct answers per hour, not tokens per second, and the two rankings do not agree.

Prats-Deck

Fifteen touch apps on a Raspberry Pi Pico 2 W: PC stats, macros, and a pet cat. C++ firmware taken from 5 to 40 frames a second with faster drawing code and the second core, and no torn frames. A simulator runs the same code on a PC, and a script taps the real deck over USB to test every app.

All ten projects, with a demo on each: prathampanchal.dev


Connect


Pinned Loading

  1. LocaLLM LocaLLM Public

    Which local model is worth running on a 12 GB card? A test harness that counts correct answers per hour.

    Python

  2. Neat Neat Public

    A Windows tray app that keeps the Downloads folder organised.

    Rust

  3. Portfolio Portfolio Public

    My portfolio: a wall of posters, one per project, each with a demo you can press.

    TypeScript

  4. Prats-Deck Prats-Deck Public

    Fifteen touch apps on a Raspberry Pi Pico 2 W: PC stats, macros, and a pet cat.

    C

  5. Scrub Scrub Public

    A local ffmpeg GUI. Trim, merge, convert and compress video with no terminal.

    TypeScript

  6. VarnikaBajpai4/FloatChat_DebugDynasty_SiH VarnikaBajpai4/FloatChat_DebugDynasty_SiH Public

    Ask questions about ocean data in plain language. Answers come from real Argo float data.

    JavaScript 1