My GitHub profile is OniiChan (@Vinayaktoor), where I describe my focus as “Building machines that learn how to learn.” I list AI/ML, reinforcement learning, and research as my areas of interest, and Delhi as my location. I have been on GitHub since 2023; the profile lists 1 follower.
One of my Python repositories is moderationdeq, a project for modeling long-term online-community equilibria under moderation policies.
- Other listed repositories include Clearcost, a tool for comparing the price of the same procedure across hospitals; Portfolio; symbolic-regression-biology, which benchmarks symbolic-regression methods on ground-truth biological models defined in SBML; -Discovery-of-hidden-symmetries-and-conservation-laws-or-Semi-supervised-Symmetry-Discovery; IA_assignment; Quantum-Reinforcement-Learning-for-High-Energy-Physics; SEAS; Vinayaktoor; projects; assignment; Unit-Converter-APP; Stopwatch-APP; c; python; and coding.
I built Clearcost, an HTML tool in my GitHub repository that compares hospital prices for the same procedure. The project parses 3 GB of real price-transparency filings from nine New York hospitals and presents the results in a self-contained web app. Its purpose is to make New York hospital prices readable; more detail is on clearcost hospital price comparison.
I have a Python repository called Neural_Network_from_Scratch, started and last updated in 2025-05. It has 1 star. The README presents it as an educational, NumPy-based implementation of a multi-layer feedforward neural network, built without TensorFlow or PyTorch.
- The network supports a custom layer-by-layer architecture through
.add(layer_size, activation). Its listed features include forward and backward propagation, ReLU and Softmax activations, one-hot encoding for multi-class classification, gradient-descent updates, accuracy reporting during training, prediction, and accuracy evaluation. The README describes the project as a way to learn neural-network training internals and build AI tools without high-level libraries.
model = Neural_Network_From_Scratch(X, Y, Iterations=1000, Alpha=0.1)
model.add(10, "relu") # hidden layer
model.add(3, "softmax") # output layer
model.compile()
The README lists the Neural_Network_From_Scratch class and its add(), compile(), predict(), and accuracy() methods, with Python 3 and NumPy as the stated technologies. It does not specify dataset details, the meaning or implementation of Alpha, the loss function, measured accuracy, input and output shapes, or installation and execution steps.
My repository descriptions include reinforcement learning for allocating capital among multiple assets to maximize long-term return and minimize risk, and for learning research strategies. SEAS is described as a self-evolving autonomous scientist that generates hypotheses, runs constrained experiments with predefined tools, evaluates results, and learns research strategies under a fixed budget. Clearcost is described as parsing 3 GB of hospital price-transparency filings from nine NYC hospitals.
I have a Python GitHub repository called RL-Based-Portfolio-Manager-Bot, started and last updated in 2025-06, with 1 star. I describe it as using Reinforcement Learning (RL) to create an intelligent agent that allocates capital among multiple assets to maximize long-term return and minimize risk. The available description does not specify the assets, algorithm, reward function, risk measures, evaluation methods, or results.
Across the listed repositories, the profile shows Python (6), Jupyter Notebook (4), HTML (2), C (1), and C++ (1).
- My contributions to other projects include 1 open pull request to google-deepmind/acme (“docs: clarify getting started entry points”); 2 merged pull requests each to Zangetsu-Tensa/LEAF-Learning-Emotions-via-Adaptive-Feedback, Zangetsu-Tensa/VisionLex-A-Multimodal-Text-Image-Understanding-Framework, and Zangetsu-Tensa/MAVNet-Multimodal-Audio-Visual-Network-for-Cross-Modal-Understanding; and 1 open pull request to sdmg15/Best-websites-a-programmer-should-visit (“Added PHP The Right Way”). The profile labels the pull requests to google-deepmind/acme and sdmg15/Best-websites-a-programmer-should-visit as open.