Karthick Aravindan

The working set

Tools I actually open.

The daily editor and models, the AI platforms around them, the charts and options tools, and the frameworks under the work: this site, Python, and running a model at the edge. This is not a ranking, and it is not every product I have tried.

Every day

These stay open. The other sections are the surrounding set, not a second daily list.

  • Editor

    Cursor (opens in a new tab)

    This is where the code gets written. Files, changes, and the agent sit in one place.

    A change is only real once it is in the project. I use Cursor for the edit itself, not as a place to collect answers.

  • Model

    Claude (opens in a new tab)

    The model I use when a problem needs a long pass: implementation, writing, and taking a system apart.

    A fluent answer is not a finished product. I still read the diff.

  • Voice

    Wispr Flow (opens in a new tab)

    I talk through the work, and Wispr Flow turns that into text I can edit.

    A lot of the first draft of a thought starts as speech. The written version is the one I keep.

  • Review

    ocra (opens in a new tab)

    I use ocra to review a change with a model, including an endpoint I choose.

    It is a review step. It is not a claim that every problem gets caught.

Models and AI platforms

Claude is in the daily set. These are the other models and platforms I actually open. This is not a ranking.

  • Models and datasets

    Hugging Face (opens in a new tab)

    The hub I use to read a model card, look at weights, and try an open model before it gets near a product.

    I use it for the model and the dataset, not as a leaderboard. A download is not an evaluation, and I am not publishing fine-tune scores here.

  • Model

    DeepSeek (opens in a new tab)

    A model I open for a second pass on code and on a hard reasoning problem.

    It sits next to Claude. I have not published a comparison, and this page does not rank them.

  • Model

    Kimi (opens in a new tab)

    Moonshot's Kimi. I use it when the material is long and I want another read of the same source.

    It is a working model for long context. It is not a published benchmark.

  • GPU platform

    NVIDIA (opens in a new tab)

    The GPU side of the AI work: the hardware, and the software stack that runs a model when an API is the wrong place for it.

    I am not naming a chip, a cluster, or a training run on this page.

Trading research

Charts and structure. These tools are for looking. Nothing here is a published strategy or a live result.

  • Charts

    TradingView (opens in a new tab)

    I use TradingView to watch price, mark a level, and keep a visual note next to the research.

    A drawing on a chart is a note. It is not a signal, and I am not publishing a setup.

  • Options structure

    StockMock (opens in a new tab)

    I use StockMock to sketch an options structure: the payoff, the margin, and what happens if price moves.

    A filled-in simulator is still a hypothesis. I have not published a return, a win rate, or the rules of a live book.

Frameworks

This site is the web stack below. Python is the model and research work. Edge computation is where a model runs when a hosted API is the wrong place. I am not listing every library.

  • Framework

    Next.js (opens in a new tab)

    The framework for this site: routes, server rendering, and the pages you are reading.

    I use it when a product is a web application and the page should be real HTML, not a shell that waits on the client.

  • UI

    React (opens in a new tab)

    The UI library under the notebook and the interactive pieces, such as the forms and the search.

    A component is a unit of the interface. It is not a reason to move work into the browser that the server can do.

  • Language

    TypeScript (opens in a new tab)

    The language for this codebase. Types are how I keep a change from drifting away from the data it claims to use.

    I am not publishing a style guide. The useful part is that a wrong shape fails before the page does.

  • Styling

    Tailwind CSS (opens in a new tab)

    The styling system for this site. The palette, the type, and the spacing live in one theme.

    Utility classes are the implementation. The visual identity is the cream, midnight, and signal system, not the framework.

  • Language

    Python (opens in a new tab)

    The language for the model work, the data, and the trading research. This website is TypeScript. That work is not.

    I use it to try a model, move data, and test a trading idea. A script that runs is not a result I am publishing.

  • Framework

    PyTorch (opens in a new tab)

    The framework I use with Python when a Hugging Face model has to run, not only sit on a model card.

    Training or inference here is a working step. I am not publishing a loss curve or a benchmark from it.

  • Where a model runs

    Edge computation (opens in a new tab)

    I use it when a model should run near the data, on a device or a private machine, instead of only as a hosted API call.

    NVIDIA, above, is the GPU platform. This entry is the placement: close to the data. I am not naming a board, a runtime, or a deployment size.