Skip to content
View Reddirector's full-sized avatar
😋
Working from home
😋
Working from home

Highlights

  • Pro

Block or report Reddirector

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. 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
Reddirector/README.md

Aditya Kumar Singh

Reddirector · Computer Science undergraduate · India

Systems engineering for AI: orchestration, reliability, agents, interpretability.

GitHub LinkedIn Portfolio Email


About

I am a second-year B.Tech CSE student who builds and ships software on my own. I care more about reliability and correctness than demos: I write tests, document known gaps, and publish releases.

I work on three things: fault-tolerant systems in Python, LLM and agent applications, and applied interpretability research. I train and test on a 6 GB RTX 3050 laptop, so I am careful about cost and efficiency.

Flagship work

Project What it is Evidence
Orchestration Engine A zero-dependency, fault-tolerant workflow engine for Python: DAG scheduling, retries with backoff, enforced timeouts, SQLite crash recovery, sagas, circuit breakers, CLI. On PyPI (v0.9.0), CI, docs. 233 tests, 91.45% coverage.*
Towards Monosemanticity Reproduction An independent, adapted reproduction of sparse-autoencoder feature learning on gelu-1l, trained on a laptop GPU. Not affiliated with Anthropic. Live dashboard. 4 validated features, 32.08x control ratio on one feature, 14/14 core tests. Limits documented.
Pentagon A bring-your-own-key AI workspace with 80+ models, local RAG, vision, voice and agentic tools, built on FastAPI, LangGraph and React. Permission gates and prompt-injection defenses for risky tool use. Web app, ~900 backend tests.* Disclosed: no multi-user auth yet.

* Self-reported from my own test suites and benchmarks.

Side projects (built with heavy AI assistance)

  • ValtSky: a Flutter client that backs up photos and videos to Telegram.
  • Megit: a Flutter music app with offline playback and synced playlists, plus a React product site.

Stack

Strong Python, pytest, mypy, GitHub Actions, SQLite, FastAPI, LangGraph, RAG (ChromaDB), MCP, React, TypeScript, Docker, Git
Learning PyTorch, TransformerLens, LoRA/QLoRA, C++ and DSA
Security testing Burp Suite, Postman, prompt-injection defenses

Currently

  • Learning PyTorch and LoRA/QLoRA fine-tuning.
  • Exploring agent crash and resume safety as the next project.
  • Light DSA practice and ongoing work on Pentagon.

Open to

Paid internships in AI infrastructure, LLM engineering, and ML research engineering. Open to relocation and remote work.


Build useful systems. Verify the result. Keep improving.

Pinned Loading

  1. Reddirector Reddirector Public

    4

  2. Orchestration_Engine Orchestration_Engine Public

    Fault-tolerant, zero-dependency workflow orchestration engine for Python: dependency-aware scheduling, retries, timeouts, SQLite-backed crash recovery.

    Python 4