A protocol for quant funds to work together without revealing their IP → the world’s first decentralized systematic fund. You can read more here.
Judge Research’s backend is low latency and can handle massively parallel processes. Ignoring our SDK and OMS for sake of simplicity, it consists of:
<aside> 👉 (users → API) + live market data providers → AWS DynamoDB → Spark + Databricks → Judge Research AI, backtesting & machine learning libraries (in R, Python and C++) → Tableau + Jupyter Notebooks + other IDEs.
</aside>
Its primary use: Users (mostly quant funds) send in features they engineer on their own backend, and our AI uses them along with our own features to construct market forecasts.
We have also been building a set of interactive dashboards to expedite (1) the analysis of our core algorithms, and (2) users’ analyses of their feature sets.
“Interactive dashboard” might sound like a pop feature. It is the frontend, however, of a quant dev framework that is profoundly disciplined - and therefore fast, flexible & scalable.
The interactive components are meant to (1) cohesively organize visually much more information than otherwise possible; or (2) facilitate the realistic counterfactual analysis of any set of models - that is, changing a setting in one chart affects the subsequent modeling & output; either as a simple ‘what if’ experiment or as a rigorously modeled counterfactual.
These dashboards have many use cases that build on one another. The below is an example.
A github repository holds a set of functions. A script calls those functions with LaTeX or markdown prose linking them. Prose, formulas and charts are laid out in the script so they mirror the final output and are therefore easily readable. The final output can be organized as an academic or publication-quality PDF, such as here, or as a Markdown webpage with interactive & live charts.
By organizing the main script as an academic article, we port all the power of our backend into an easy-to-read format that saves considerable work hours while - for example - engaging with clients about the particulars of a pricing formula, or onboarding new employees.
By extending the interactive functions of the dashboard, we begin asking practical questions that can govern trading decisions, for either a quant dev or a trader with no coding skills.
The goal is a single, customized dashboard that controls each step of the analysis and reruns the subsequent steps upon their change. This gives traders the ability to make fully modeled, informed decisions with a single click instead of hours of dev time.