Simple Things

machine learning experiments, competitive programming solutions, coding concepts, and tutorials

Repository Structure

CP (Competitive Programming)

solutions and templates for various algorithmic platforms:

• Codeforces: Problem solutions mainly in C++.
• CodeChef: Problem solutions.
• AtCoder: Contest solutions.
• CSES: Standard problem set solutions.
• Template: Reusable code snippets and templates for competitive programming.

Coding

Notes and examples on core Computer Science topics:

• DB: Database concepts and implementations.
• Design Pattern: Common software design patterns.
• OS: Operating System internal concepts.
• System Design: Scalable system architecture notes.

Interpretability

Research and experiments in Mechanistic Interpretability:

• Manipulation: Studies on model manipulation and steering (e.g., Chain of Thought).
• Lens: Tools and visualizations for inspecting model internals.

ML (Machine Learning)

Implementations of fundamental ML components from scratch:

• MLP: Multi-Layer Perceptron implementation.
• Attention: Self-attention interaction mechanisms.
• RoPE: Rotary Positional Embeddings.
• KV Cache: Key-Value caching for efficient inference.
• Parallelism: Model and data parallelism concepts.
• Solutions: Reference implementations for the exercises.

ARENA 3.0

The Alignment Research Engineer Accelerator (ARENA) curriculum, focusing on:

• Chapter 0: Fundamentals: Core concepts and prerequisites.
• Chapter 1: Transformer Interpretability: Deep dive into transformer circuits and interpretation.
• Chapter 2: Reinforcement Learning: RL algorithms and their applications.
• Chapter 3: LLM Evals: Techniques for evaluating Large Language Models.

Alogverse

Notebooks and challenges related to Algoverse:

• Bhagyesh Kumar Evals Challenge.ipynb
• Bhagyesh Kumar Interpretability Challenge copy.ipynb
• misc.ipynb

Tutorial

Project-based tutorials and practical implementations:

• Face-landmark: Facial landmark detection systems.
• Fake-News-detection: NLP models for classifying fake news.
• GAN: Generative Adversarial Networks experiments.
• tic-tac: Game development and AI implementations for Tic-Tac-Toe.

---

External Projects List

Ongoing and past projects maintained continuously:

Project	Description
image-studio	Modular image processing tool and studio.
TopoReformer	Research on adversarial robustness via Persistence Homology.
LM-quant	Language model quantization techniques.

Documentation

• status.md: Current project status and progress tracking.
• links.md: Repository of useful resources and references.