Skip to content
View Agarwalrishu13's full-sized avatar
4 followers · 4 following

Highlights

  • Pro

Block or report Agarwalrishu13

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 don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
Agarwalrishu13/README.md

Priyanshu Agarwal

Software developer focused on local-first software and AI built from scratch — tools that run entirely on your own machine, with no accounts, no telemetry and nothing to install. B.Tech student in India, shipping real projects since 2024.

The nano family — 11 apps for people who don't code

A family of offline-first desktop apps that share one shape: a double-click to start, a web page instead of a command line, the Python standard library instead of dependencies, and plain words everywhere. 800+ automated tests across the repos.

project what it does
🗺 nano the family map — all apps, how they fit together
🧠 nanoLaama chat with a local AI, fully offline
📚 nanoDoc ask a document questions, with the source paragraph shown
🗂 nanoGit version control without learning git — 218 tests
🧲 nanoPick find files by describing what you remember

AI from the transistors up

Not calling APIs — building the stack underneath them:

  • ⚙️ nanollama.c — an LLM inference engine in ~1,400 lines of dependency-free C: transformer forward pass, BPE tokenizer, hand-rolled thread pool, int8 quantization. No BLAS, no OpenMP, no API keys.
  • 🧬 nanobrain — a Llama-2 transformer trained from scratch in raw PyTorch (16.9M params, 35 hours on CPU), exported to my own engine's binary format.
  • 🏗 nanoforge · 🎯 nanorl — the studio and the SFT/DPO alignment trainers that complete the pipeline.

Selected work

  • MediaForge — drag-and-drop AI media builder: visual node editor (ReactFlow) over a FastAPI backend for image/video/audio models.
  • finguard — financial fraud detection platform: FastAPI + scikit-learn + React 18, with a hybrid explainable detection engine.
  • trek-dtn-messenger — delay-tolerant messaging agent for intermittent networks via greedy SMTP flushing.

Stack
Languages Python · C · TypeScript / JavaScript · Java
AI / ML PyTorch (training from scratch), scikit-learn, SFT / DPO, transformers
Web FastAPI, React, Next.js, Node.js, PostgreSQL, MongoDB
Practice GitHub Actions CI on every repo, unittest suites, stdlib-first design

Contact

Pinned Loading

  1. nanollama.c nanollama.c Public

    A complete LLM inference engine from scratch in ~1400 lines of dependency-free C — transformer, BPE tokenizer, sampler, hand-rolled thread pool, int8 quantization. No OpenMP, no BLAS, no API keys.

    C 2 1

  2. nanobrain nanobrain Public

    I trained my own LLM from scratch — hand-written Llama-2 transformer, my own BPE tokenizer, my own exporter — and it runs on nanollama.c, the C inference engine I also built. No APIs. TinyStories. …

    Python 1

  3. nanoforge nanoforge Public

    Python 1

  4. HostelLaundry HostelLaundry Public

    Android app for hostel students to book washing machine slots (Java, SQLite, Material 3) - MAD project

    Java 1

  5. finguard finguard Public

    AI-powered financial fraud detection & risk intelligence platform - FastAPI + scikit-learn + React 18 (Tailwind v4), with SRS, test suite, and a hybrid explainable detection engine

    TypeScript 1

  6. trek-dtn-messenger trek-dtn-messenger Public

    Asymmetric delay-tolerant messaging agent utilizing greedy SMTP flushing for intermittent network environments

    Python 1