Jane Street Feature Trails

Feature trailActive

Intern Projects Showcase

This post showcases a variety of projects completed by Jane Street interns across different engineering and research domains, highlighting their contributions to areas like ML research, software engineering, Linux engineering, and trading desk operations. This edition specifically details the development of a centralized configuration management orchestrator for Windows, replacing a distributed PowerShell system with a more robust OCaml and F# based solution, including an audit mode for graceful migration and discrepancy detection.

4 posts · 2026

Timeline

Feature trailActive

Messaging

This post extends the messaging capability by detailing optimizations to Aria, an internal messaging framework. It introduces a new indexing strategy for tip recovery, reducing CPU usage by 30% and addressing server overload. Additionally, it details an adaptive splitting algorithm for subtree stores to improve initial recovery performance by intelligently distributing messages based on volume, mitigating issues with filtering excess data and allowing users to structure topics more intuitively. The work involved benchmarking, using AI agents for heap implementation selection and UI visualization, and implementing a greedy clustering algorithm with binary search for adaptive splitting.

1 post · 2026

Timeline

Feature trailActive

ASIC Reverse Engineering Puzzle

This post extends the ASIC reverse engineering puzzle capability by revealing the solution to the puzzle and detailing the various methods solvers used to reverse-engineer the chip. It highlights techniques such as netlist extraction from GDS layout, simulation using cell models or custom evaluators, logical analysis of circuit modules, dynamic signal exploration, and SAT solving. The post also details the chip's functionality as a hardware checker for an 11x11 Star Battle puzzle and discusses the obfuscation used in the output generator.

2 posts · 2026

Timeline

Feature trailActive

Linux Kernel Bug Hunting

This post extends the Linux kernel bug hunting capability by detailing the discovery and resolution of six bugs found during recurring network maintenance. These bugs affected glibc, the OCaml networking stack, Async, the Kafka client, and a job runner, highlighting issues with DNS resolution, socket handling, request retries, and file descriptor limits. The post details the investigation process, the root causes of each bug (including an eleven-year-old glibc bug and a fourteen-year-old Async regression), and the fixes implemented. It also discusses the interplay between different system layers and the importance of robust error handling and resource management in distributed systems.

4 posts · 2017–2026

Timeline

20172026

Feature trailActive

Activation Checkpointing

This post introduces a new capability for activation checkpointing in model training. It details a novel procedure for finding optimal nodes to checkpoint to trade off compute for memory, addressing limitations in existing PyTorch tools. The procedure involves a planning pass to estimate true memory usage and recomputation costs, followed by a greedy algorithm to select nodes for checkpointing based on the cheapest recompute cost per byte of peak memory freed. This capability aims to provide use cases for autoregressive diffusion models in quantitative finance, specifically for generating market data. The post details the challenges of modeling market data, which exhibits both continuous and discrete characteristics, and explores solutions using diffusion and flow-matching models. It highlights findings on the instability of DDPM and the better performance of flow matching. The post also discusses techniques for handling discontinuities in market data, such as fracturing categorical predictions and using 'atom smoothing'. Finally, it presents results comparing model-generated data to real market data using total variational divergence and a discriminator classifier.

3 posts · 2026

Timeline

Feature trailActive

Sequence Weighting at Scale

This post introduces a new capability for studying sequence weighting at scale, detailing the technical approach and its benefits for improving model performance and efficiency. This post extends that capability by exploring novel methods for sequence weighting, potentially leading to more efficient and accurate models.

1 post · 2026

Timeline

Feature trailActive

Hardware Design Testing

This post extends the hardware design testing discussion by introducing a competition to design an open-source, general-purpose protocol emulator ASIC. It details the technical approach of designing a reprogrammable chip to support new protocols after fabrication, aiming to foster skills in hardware analysis and design exploration. This capability extends to adapting the programming language to better support proof-oriented techniques and integrating with external formal methods tools.

8 posts · 2020–2026

Timeline

20202026

Feature trailActive

OCaml Build System

This post extends the OCaml build system discussion by detailing how an accidental discovery led to a more robust and efficient build system. It highlights the technical challenges and solutions encountered in integrating various components and optimizing build times, contributing to the broader development of versatile OCaml systems. This post specifically addresses a versioning problem within the ppx preprocessor, proposing a solution that simplifies dependency management and improves build reliability.

7 posts · 2016–2026

Timeline

…20172026

Feature trailActive

Formal Methods Integration

This post introduces a new capability for integrating formal methods into software development, driven by the emergence of agentic coding. It details the technical approach to leveraging formal methods to address the verification bottleneck and provide feedback to AI agents, aiming to make formal methods as pervasively useful as type systems. This capability extends to adapting the programming language to better support proof-oriented techniques and integrating with external formal methods tools.

1 post · 2026

Timeline

Feature trailActive

TUI Development and Debugging Tools

This post extends the TUI development and debugging tools capability by introducing strace-ui and Bonsai_term. These tools enhance the development and debugging experience for terminal user interfaces (TUIs) by providing better visibility and control over TUI applications, contributing to the broader renaissance of TUI development.

1 post · 2026

Timeline

Feature trailActive

Positional Encoding Exploration for Attention Models

This post introduces a novel approach to exploring positional encodings for attention mechanisms by leveraging group theory. It details the mathematical framework and its application in generating and analyzing different encoding spaces, aiming to improve the understanding and performance of attention-based models. This post extends that capability by exploring novel methods for sequence weighting, potentially leading to more efficient and accurate models.

1 post · 2026

Timeline

Feature trailQuiet since 2024

Piecewise Linear Neural Network Visualization

This post extends the capability for visualizing piecewise linear neural networks by introducing a novel approach to visualizing the decision boundaries of these networks. It details the technical approach of using a combination of techniques to generate and interpret these visualizations, providing deeper insights into the decision boundaries and internal workings of such networks.

1 post · 2024

Timeline

20242026

Feature trailQuiet since 2023

Reproducible Python Environments with XARs

This post introduces a new capability for building reproducible Python environments using XARs, detailing the technical approach and its benefits.

1 post · 2023

Timeline

20232026

Feature trailQuiet since 2022

Performance Diagnosis with Intel Processor Trace

This post introduces Magic-trace, a new capability for diagnosing tricky performance issues using Intel Processor Trace. It details the technical approach and its benefits for performance engineering.

1 post · 2022

Timeline

20222026

Feature trailQuiet since 2020

Memory Leak Detection with Memtrace

This post introduces Memtrace, a new capability for finding memory leaks in OCaml programs. It details the technical approach of using heap profiling to identify and diagnose memory leaks, contributing to improved program stability and resource management.

1 post · 2020

Timeline

20202026

Feature trailQuiet since 2020

Systemd Troubleshooting with SystemTap

This post introduces a new capability for troubleshooting systemd using SystemTap, detailing the technical approach and its benefits.

1 post · 2020

Timeline

20202026

Feature trailQuiet since 2019

Deep Learning for Go Problem Solving

This post introduces a new capability for applying deep learning techniques to solve complex problems in the game of Go. It details the technical approach of using neural networks to analyze game states and predict optimal moves, aiming to advance AI capabilities in strategic games.

1 post · 2019

Timeline

20192026

Feature trailQuiet since 2019

Self-Play Learning Acceleration

This post introduces a new capability for accelerating self-play learning in Go, detailing the technical approach and its benefits.

1 post · 2019

Timeline

20192026

Feature trailQuiet since 2019

Deep Reinforcement Learning for Atari Games

This post introduces a new capability for playing Atari games using OCaml and deep reinforcement learning. It details the technical approach of applying deep learning techniques to game states and predicting optimal moves, aiming to advance AI capabilities in game playing.

1 post · 2019

Timeline

20192026

Feature trailQuiet since 2018

Automated Theorem Proving with Z3

This post introduces a new capability for automated theorem proving using the Z3 SMT solver. It details the technical approach to leveraging Z3 for generating proofs and refutations, contributing to formal verification and software correctness.

1 post · 2018

Timeline

20182026

Feature trailQuiet since 2017

Network Data Plane Verification

This post introduces a new capability for verifying network data planes. It details the technical approach to ensuring the correctness and reliability of network infrastructure by systematically testing its data plane operations.

1 post · 2017

Timeline

20172026

Feature trailQuiet since 2017

Tree Diffing Algorithms

This post introduces a new capability for designing and implementing tree diffing algorithms, detailing the technical approach and its benefits for comparing hierarchical data structures.

1 post · 2017

Timeline

20172026