Machine Learning Researcher at Kronos Research

This is a Quantitative Research role with a heavy Engineering twist. Kronos is a high-frequency trading (HFT) and market-making firm, heavily focused on crypto. Unlike a data scientist at a tech company who optimizes for accuracy (AUC/RMSE), you are optimizing for Latency-Adjusted Alpha.

Your job is to build Deep Learning models (Transformers, RNNs) that can predict price movements based on Orderbook Data, but—crucially—these models must run fast enough to trade in real-time. If your model is 99% accurate but takes 500ms to run, it is useless because the market has already moved.

## Key Responsibilities
The "Alpha" Hunter: You will design neural networks (Transformers, CNNs, Attention mechanisms) to find patterns in market noise.

The Data Plumber: You are working with Level 2/Level 3 Orderbook data. This is high-granularity data showing every bid and ask. You must preprocess this using C++ because Python is too slow for the ingestion pipeline.

The Optimizer: The JD explicitly mentions "KV caching, quantization, pruning." This means you need to take a massive Transformer model and shrink it/optimize it so it runs on inference servers in microseconds.

Deployment: You don't just hand the model to a developer. You are expected to integrate it into the C++ production trading system yourself.

## Strategic Analysis
The "Fresh Grad" Opportunity: Unusually for a top-tier prop shop, they are open to "highly motivated fresh graduates." However, the application form asks for GPA, Transcripts, and Academic Awards. They are looking for raw IQ and mathematical aptitude over industry experience.

The Tech Stack Signal: They mention JAX. JAX is increasingly popular in high-end research for its ability to compile high-performance code (via XLA) and handle complex gradients. If you know JAX, you have a massive advantage.

The Crypto Context: Kronos is a giant in crypto market making. The crypto markets run 24/7 and are incredibly volatile. Your models need to be robust enough not to blow up the account during a "flash crash."

## Candidate Profile
The "Type A" Student: If you are a fresh grad, you likely have a 3.9+ GPA, Olympiad medals, or published papers. They are screening for elite academic pedigree.

The Hybrid Engineer: You are comfortable reading a research paper on Attention Is All You Need, implementing it in PyTorch/JAX, and then writing a C++ wrapper to deploy it.

The Latency Architect: You understand computer architecture. You know why a cache miss is expensive and how floating-point precision (FP32 vs INT8) impacts speed.

## Critical Application "Knockout" Factors
C++ Proficiency: This is a hard requirement. If you can only code in Python, you will likely fail the technical screen.

Academic Excellence: For candidates with <3 years experience, the "Transcript" and "GPA" fields are likely the primary filters.

Orderbook Knowledge: Understanding how an Orderbook works (Bid/Ask spread, depth, matching engine mechanics) is crucial.

## Next Step
The interview will likely involve a mix of Math Puzzles, Machine Learning Theory, and C++ Coding.