# Jane Street

OCaml and Systems rolled into one

This is one page of public article previews, not the complete archive. Follow Next page to continue. Summaries are not the original full articles.

## Writing a DSL for high performance parallel programming in OxCaml

DevFeed: [Writing a DSL for high performance parallel programming in OxCaml](<https://devfeed.tech/articles/writing-a-dsl-for-high-performance-parallel-programming-in-oxcaml-132544.md>)

Original publisher: [Read original article](<https://blog.janestreet.com/writing-a-dsl-for-high-performance-parallel-programming-in-oxcaml/>)

Author: Nicholas Yang

Published: 2026-11-06T00:00:00Z

Content type: article

Language: en

Sources: [Jane Street](<https://devfeed.tech/sources/jane-street.md>)

Topics: [High-Performance Computing](<https://devfeed.tech/topics/high-performance-computing.md>), [vectorization techniques](<https://devfeed.tech/topics/vectorization-techniques.md>), [Programming](<https://devfeed.tech/topics/programming.md>), [conditionals](<https://devfeed.tech/topics/conditionals.md>), [Test automation](<https://devfeed.tech/topics/test-automation.md>), [iteration](<https://devfeed.tech/topics/iteration.md>), [GPU](<https://devfeed.tech/topics/gpu.md>)

Tags: [2026](<https://devfeed.tech/tags/2026.md>), [algorithm](<https://devfeed.tech/tags/algorithm.md>), [architecture](<https://devfeed.tech/tags/architecture.md>), [architectures](<https://devfeed.tech/tags/architectures.md>), [arithmetic](<https://devfeed.tech/tags/arithmetic.md>), [arrays](<https://devfeed.tech/tags/arrays.md>), [assembly](<https://devfeed.tech/tags/assembly.md>), [avx2](<https://devfeed.tech/tags/avx2.md>), [avx512](<https://devfeed.tech/tags/avx512.md>), [backend](<https://devfeed.tech/tags/backend.md>), [benchmarking](<https://devfeed.tech/tags/benchmarking.md>), [benchmarks](<https://devfeed.tech/tags/benchmarks.md>), [c](<https://devfeed.tech/tags/c.md>), [dsl](<https://devfeed.tech/tags/dsl.md>), [high-performance](<https://devfeed.tech/tags/high-performance.md>), [intern](<https://devfeed.tech/tags/intern.md>), [internship](<https://devfeed.tech/tags/internship.md>), [parallel](<https://devfeed.tech/tags/parallel.md>), [performance](<https://devfeed.tech/tags/performance.md>), [programming](<https://devfeed.tech/tags/programming.md>), [writing](<https://devfeed.tech/tags/writing.md>)

### AI overview

The OxCaml DSL lets developers describe the computation for one output index in an SPMD kernel; its compiler builds a graph and uses runtime metaprogramming to specialize and compile OxCaml code for the chosen metric and target architecture. Kernels write each output entry once, enabling parallel execution. The compiler distinguishes uniform and per-lane values, and offers `if_`, `if_var` and `select`; unlike `select`, which evaluates both branches, `if_var` can skip an untaken branch. An A-normal-form IR makes transformations such as hoisting loop-invariant loads and processing four elements per loop iteration easier to represent and optimize. The initial JIT was much faster than the interpreter, about as fast as or slightly slower than manual OxCaml, and significantly slower than C AVX2/AVX512. The IR backend initially performed about the same as that JIT. Preliminary vectorization and constant-hoisting work brought results on par with C AVX2; scheduling improvements were still in progress, and the excerpt reports no final results for them.

### Source excerpt

The following is part of a series of posts about 2026 summer intern projects--for more, see "What the interns have wrought, special jumbo 2026 edition"

## Our software engineering interviews are changing (a little)

DevFeed: [Our software engineering interviews are changing (a little)](<https://devfeed.tech/articles/our-software-engineering-interviews-are-changing-a-little-122087.md>)

Original publisher: [Read original article](<https://blog.janestreet.com/software-engineering-interviews-are-changing-a-little/>)

Author: Matthew Saffer

Published: 2026-11-02T00:00:00Z

Content type: article

Language: en

Sources: [Jane Street](<https://devfeed.tech/sources/jane-street.md>)

Topics: [Software Engineering](<https://devfeed.tech/topics/software-engineering.md>), [AI-assisted coding](<https://devfeed.tech/topics/ai-assisted-coding.md>), [engineering-culture](<https://devfeed.tech/topics/engineering-culture.md>)

Tags: [coding](<https://devfeed.tech/tags/coding.md>), [design](<https://devfeed.tech/tags/design.md>), [formal-methods](<https://devfeed.tech/tags/formal-methods.md>), [interviews](<https://devfeed.tech/tags/interviews.md>), [llms](<https://devfeed.tech/tags/llms.md>), [process](<https://devfeed.tech/tags/process.md>), [software-engineering](<https://devfeed.tech/tags/software-engineering.md>)

### AI overview

Jane Street is trialing some new onsite interview rounds for campus candidates that involve less code-writing and focus more on software design, checking code quality, structure and intent, and using LLMs to produce code. Most rounds remain traditional coding interviews without LLM assistance, and not every onsite includes the new formats yet; candidates will be told what to expect. The broader process is unchanged, and candidates need no specialized agent workflow or particular LLM setup. The article says interviewers value clear judgment about tools' strengths, limits, and when to use alternatives. More experienced candidates will continue to see varied interview styles.

### Source excerpt

Over the last few years, LLMs have changed how software gets built and that's no different for us at Jane Street. Most of our engineers now use them every day not only for writing code, but also for reading, debugging, and reviewing code, and plenty of things that aren't code at all.

## Flex Model: an event-driven feature engineering system

DevFeed: [Flex Model: an event-driven feature engineering system](<https://devfeed.tech/articles/flex-model-an-event-driven-feature-engineering-system-88874.md>)

Original publisher: [Read original article](<https://blog.janestreet.com/flex-model-an-event-driven-feature-engineering-system/>)

Author: Stanislaw Maczka

Published: 2026-10-13T00:00:00Z

Content type: article

Language: en

Sources: [Jane Street](<https://devfeed.tech/sources/jane-street.md>)

Topics: [Feature Engineering](<https://devfeed.tech/topics/feature-engineering.md>), [data-engineering](<https://devfeed.tech/topics/data-engineering.md>), [event driven](<https://devfeed.tech/topics/event-driven.md>), [Functional programming](<https://devfeed.tech/topics/functional-programming.md>)

Tags: [engineering](<https://devfeed.tech/tags/engineering.md>), [event-driven](<https://devfeed.tech/tags/event-driven.md>), [feature](<https://devfeed.tech/tags/feature.md>), [feature-engineering](<https://devfeed.tech/tags/feature-engineering.md>)

### AI overview

Jane Street describes Flex Model, an event-driven feature engineering framework that converts market events into numerical features for trading systems. The article explains its stateful feature model, a planner that schedules handlers and evaluations for different event types, and performance work including preallocated memory and OxCaml unboxed floats. The system aims to let researchers compose reusable features while limiting unnecessary computation across markets.

### Source excerpt

Neural networks are a tool we use heavily in many trading strategies. But raw market data can't be fed to a neural net as-is. It's an unstructured input that needs normalization and feature extraction first.

## Building and testing a config management orchestrator for Windows

DevFeed: [Building and testing a config management orchestrator for Windows](<https://devfeed.tech/articles/building-and-testing-a-config-management-orchestrator-for-windows-65867.md>)

Original publisher: [Read original article](<https://blog.janestreet.com/config-management-orchestrator-for-windows/>)

Author: Israel Milgrom

Published: 2026-10-06T00:00:00Z

Content type: article

Language: en

Sources: [Jane Street](<https://devfeed.tech/sources/jane-street.md>)

Topics: [configuration-management](<https://devfeed.tech/topics/configuration-management.md>), [CI/CD](<https://devfeed.tech/topics/cicd.md>), [Continuous Delivery (CD)](<https://devfeed.tech/topics/continuous-delivery.md>), [developer setup](<https://devfeed.tech/topics/developer-setup.md>)

Tags: [audit](<https://devfeed.tech/tags/audit.md>), [configuration-management](<https://devfeed.tech/tags/configuration-management.md>), [engineering](<https://devfeed.tech/tags/engineering.md>), [firewall](<https://devfeed.tech/tags/firewall.md>), [internship](<https://devfeed.tech/tags/internship.md>), [windows](<https://devfeed.tech/tags/windows.md>)

### AI overview

The article describes replacing a per-host PowerShell configuration management process for Windows with a centralized Orchestrator service and a stateless F# client. An OCaml parser handles applicability, scheduling and rollout phasing, while audit mode compares the new system's decisions with the old system to reveal discrepancies and bugs. The migration is proceeding gradually, with the previous monitoring retained until the new setup is proven.

### Source excerpt

The following is part of a series of posts about 2026 summer intern projects--for more, see "What the interns have wrought, special jumbo 2026 edition"

## Scaling and benchmarking a critical message bus using a new indexing strategy

DevFeed: [Scaling and benchmarking a critical message bus using a new indexing strategy](<https://devfeed.tech/articles/scaling-and-benchmarking-a-critical-message-bus-using-a-new-indexing-strategy-65868.md>)

Original publisher: [Read original article](<https://blog.janestreet.com/scaling-and-benchmarking-a-critical-message-bus/>)

Author: Nicholas Yang

Published: 2026-10-05T00:00:00Z

Content type: article

Language: en

Sources: [Jane Street](<https://devfeed.tech/sources/jane-street.md>)

Topics: [Publish-subscribe pattern](<https://devfeed.tech/topics/pubsub.md>), [clustering](<https://devfeed.tech/topics/clustering.md>), [Data Mining & Modeling](<https://devfeed.tech/topics/data-mining-modeling.md>)

Tags: [benchmarking](<https://devfeed.tech/tags/benchmarking.md>), [clustering](<https://devfeed.tech/tags/clustering.md>), [cpu-utilization](<https://devfeed.tech/tags/cpu-utilization.md>), [data-structure](<https://devfeed.tech/tags/data-structure.md>), [internship](<https://devfeed.tech/tags/internship.md>), [scaling](<https://devfeed.tech/tags/scaling.md>)

### AI overview

An intern's work on Jane Street's Aria messaging system improved tip recovery by indexing messages by topic partition and tuning the index with benchmarks. The post also describes an adaptive topic-tree splitting algorithm intended to reduce unnecessary reads during initial recovery. Indexing has shipped to production, where staging results showed 30% lower CPU utilization in some scenarios; adaptive splitting is in staging.

### Source excerpt

The following is part of a series of posts about 2026 summer intern projects--for more, see "What the interns have wrought, special jumbo 2026 edition"

## How recurring network maintenance exposed 6 bugs

DevFeed: [How recurring network maintenance exposed 6 bugs](<https://devfeed.tech/articles/how-recurring-network-maintenance-exposed-6-bugs-64210.md>)

Original publisher: [Read original article](<https://blog.janestreet.com/how-recurring-network-maintenance-exposed-6-bugs/>)

Author: Adam Yi

Published: 2026-10-02T00:00:00Z

Content type: article

Language: en

Sources: [Jane Street](<https://devfeed.tech/sources/jane-street.md>)

Topics: [Software Engineering](<https://devfeed.tech/topics/software-engineering.md>), [Application Development](<https://devfeed.tech/topics/application-development.md>), [distributed algorithm](<https://devfeed.tech/topics/distributed-algorithm.md>), [Load Balancing](<https://devfeed.tech/topics/load-balancing.md>)

Tags: [bugs](<https://devfeed.tech/tags/bugs.md>), [infrastructure](<https://devfeed.tech/tags/infrastructure.md>), [kafka](<https://devfeed.tech/tags/kafka.md>), [network](<https://devfeed.tech/tags/network.md>), [sockets](<https://devfeed.tech/tags/sockets.md>), [tcp](<https://devfeed.tech/tags/tcp.md>)

### AI overview

Recurring network partitions at Jane Street triggered failures in its Kafka infrastructure, including a glibc resolver crash, a UDP socket leak, and a surge of TCP connections during recovery. The investigation traced the incidents to interacting bugs in glibc, OCaml networking and Async code, Kafka client retries and bootstrapping, and a job-runner migration that lowered file-descriptor limits. Fixes addressed each layer, and the article draws lessons about timeout cancellation, failure-atomic setup, shared resources, and testing failure paths.

### Source excerpt

Jane Street does a lot of network maintenance over the weekends. In each maintenance, a cabinet may lose connectivity for 15 to 25 minutes. Our internal Kafka infrastructure is supposed to be resilient to these partitions, and reconnect when connectivity resumes. For several months this year, it didn't: processes segfaulted, leaked hundreds of thousands of sockets, or crashed in a surge of half a million TCP connections.

## Results from the ASIC puzzle

DevFeed: [Results from the ASIC puzzle](<https://devfeed.tech/articles/results-from-the-asic-puzzle-63798.md>)

Original publisher: [Read original article](<https://blog.janestreet.com/asic-puzzle-results/>)

Author: Benjamin Devlin

Published: 2026-10-02T00:00:00Z

Content type: article

Language: en

Sources: [Jane Street](<https://devfeed.tech/sources/jane-street.md>)

Topics: [Chip design](<https://devfeed.tech/topics/chip-design.md>), [Reverse Engineering](<https://devfeed.tech/topics/reverse-engineering.md>), [code analysis](<https://devfeed.tech/topics/code-analysis.md>), [reproducibility](<https://devfeed.tech/topics/reproducibility.md>)

Tags: [a-star](<https://devfeed.tech/tags/a-star.md>), [asic](<https://devfeed.tech/tags/asic.md>), [bug](<https://devfeed.tech/tags/bug.md>), [c-plus-plus](<https://devfeed.tech/tags/c-plus-plus.md>), [challenges](<https://devfeed.tech/tags/challenges.md>), [checksums](<https://devfeed.tech/tags/checksums.md>), [internals](<https://devfeed.tech/tags/internals.md>), [reverse](<https://devfeed.tech/tags/reverse.md>), [solved](<https://devfeed.tech/tags/solved.md>), [up](<https://devfeed.tech/tags/up.md>)

### AI overview

Jane Street reports the results of a reverse-engineering puzzle built around an ASIC that checks solutions to an 11x11 Star Battle puzzle. Entrants used netlist extraction, simulation, logic analysis, SAT solving, and other methods to uncover the circuit's behavior. The article also highlights debugging lessons, including how testing only against a supplied waveform can leave modeling errors undetected, and discusses uses of AI in building analysis and debugging tools.

### Source excerpt

In August, we published a puzzle that handed you the final GDS layout of a small chip and asked you to work out what it did. We supplied the physical layout, but not a netlist or internal signal names, and it was up to you to reverse-engineer the internals. This post reveals what the chip does and explores the different ways you solved it, with shout-outs to some of our favorite submissions.

## A FUSE server for our in-house storage backend

DevFeed: [A FUSE server for our in-house storage backend](<https://devfeed.tech/articles/a-fuse-server-for-our-in-house-storage-backend-63555.md>)

Original publisher: [Read original article](<https://blog.janestreet.com/linux-engineering-depot-fuse-index/>)

Author: Ben Steffen

Published: 2026-10-01T00:00:00Z

Content type: article

Language: en

Sources: [Jane Street](<https://devfeed.tech/sources/jane-street.md>)

Topics: [systems](<https://devfeed.tech/topics/systems.md>), [Magic Pocket](<https://devfeed.tech/topics/magic-pocket.md>), [Back end](<https://devfeed.tech/topics/backend.md>)

Tags: [allows](<https://devfeed.tech/tags/allows.md>), [async](<https://devfeed.tech/tags/async.md>), [backend](<https://devfeed.tech/tags/backend.md>), [benchmarks](<https://devfeed.tech/tags/benchmarks.md>), [blocking](<https://devfeed.tech/tags/blocking.md>), [bug](<https://devfeed.tech/tags/bug.md>), [capture](<https://devfeed.tech/tags/capture.md>), [decoupling](<https://devfeed.tech/tags/decoupling.md>), [errors](<https://devfeed.tech/tags/errors.md>), [faster](<https://devfeed.tech/tags/faster.md>), [file](<https://devfeed.tech/tags/file.md>), [internship](<https://devfeed.tech/tags/internship.md>), [series](<https://devfeed.tech/tags/series.md>), [server](<https://devfeed.tech/tags/server.md>), [storage](<https://devfeed.tech/tags/storage.md>)

### AI overview

Jane Street describes an OCaml-native FUSE library and read-only server that exposes its Depot object store through a POSIX filesystem interface. The article covers protocol implementation, compatibility decisions for Depot paths and names, and performance work that raised large-file read speeds from about 30 MiB/s to 1.3 GiB/s.

### Source excerpt

The following is part of a series of posts about 2026 summer intern projects--for more, see "What the interns have wrought, special jumbo 2026 edition"

## A resilient kernel log reporter

DevFeed: [A resilient kernel log reporter](<https://devfeed.tech/articles/a-resilient-kernel-log-reporter-63556.md>)

Original publisher: [Read original article](<https://blog.janestreet.com/linux-engineering-kernel-reporter-index/>)

Author: Varun Gupta

Published: 2026-10-01T00:00:00Z

Content type: article

Language: en

Sources: [Jane Street](<https://devfeed.tech/sources/jane-street.md>)

Topics: [Kernel](<https://devfeed.tech/topics/kernel.md>), [Linux](<https://devfeed.tech/topics/linux.md>), [log management](<https://devfeed.tech/topics/log-management.md>), [Cloud Logging](<https://devfeed.tech/topics/cloud-logging.md>)

Tags: [analysis](<https://devfeed.tech/tags/analysis.md>), [backend](<https://devfeed.tech/tags/backend.md>), [backpressure](<https://devfeed.tech/tags/backpressure.md>), [capture](<https://devfeed.tech/tags/capture.md>), [cause](<https://devfeed.tech/tags/cause.md>), [chrome](<https://devfeed.tech/tags/chrome.md>), [cli](<https://devfeed.tech/tags/cli.md>), [collection](<https://devfeed.tech/tags/collection.md>), [debug](<https://devfeed.tech/tags/debug.md>), [device](<https://devfeed.tech/tags/device.md>), [dns](<https://devfeed.tech/tags/dns.md>), [file](<https://devfeed.tech/tags/file.md>), [filesystem](<https://devfeed.tech/tags/filesystem.md>), [internship](<https://devfeed.tech/tags/internship.md>), [kernel](<https://devfeed.tech/tags/kernel.md>)

### AI overview

A Linux log shipper collects kernel logs without depending on local disk or network availability. The project also uses pstore to preserve panic logs for transfer after reboot, then exposes them through command-line tools to help diagnose crashes, including failures associated with a recent Nvidia driver update.

### Source excerpt

The following is part of a series of posts about 2026 summer intern projects--for more, see "What the interns have wrought, special jumbo 2026 edition"

## Trading off compute for memory with activation checkpointing

DevFeed: [Trading off compute for memory with activation checkpointing](<https://devfeed.tech/articles/trading-off-compute-for-memory-with-activation-checkpointing-62850.md>)

Original publisher: [Read original article](<https://blog.janestreet.com/trading-off-compute-for-memory-with-activation-checkpointing/>)

Author: James Somers

Published: 2026-09-30T00:00:00Z

Content type: article

Language: en

Sources: [Jane Street](<https://devfeed.tech/sources/jane-street.md>)

Topics: [memory handling](<https://devfeed.tech/topics/memory-handling.md>)

Tags: [compute](<https://devfeed.tech/tags/compute.md>), [internship](<https://devfeed.tech/tags/internship.md>), [memory](<https://devfeed.tech/tags/memory.md>), [model-training](<https://devfeed.tech/tags/model-training.md>), [pytorch](<https://devfeed.tech/tags/pytorch.md>)

### AI overview

Activation checkpointing reduces model-training memory use by discarding selected activations and recomputing them during the backward pass. The article introduces an internship project to choose which activations to discard under a memory budget while minimizing extra computation, and examines PyTorch's existing mechanisms.

### Source excerpt

The following is part of a series of posts about 2026 summer intern projects--for more, see "What the interns have wrought, special jumbo 2026 edition"

## Mitigating memorization in LLMs

DevFeed: [Mitigating memorization in LLMs](<https://devfeed.tech/articles/mitigating-memorization-in-llms-62848.md>)

Original publisher: [Read original article](<https://blog.janestreet.com/mitigating-memorization-in-llms/>)

Author: James Somers

Published: 2026-09-30T00:00:00Z

Content type: article

Language: en

Sources: [Jane Street](<https://devfeed.tech/sources/jane-street.md>)

Topics: [Large Language Model](<https://devfeed.tech/topics/llm.md>), [Occam's razor machine learning](<https://devfeed.tech/topics/occam-s-razor-machine-learning.md>), [qwen](<https://devfeed.tech/topics/qwen.md>), [autonomous vehicles](<https://devfeed.tech/topics/autonomous-vehicles.md>)

Tags: [internship](<https://devfeed.tech/tags/internship.md>), [llms](<https://devfeed.tech/tags/llms.md>), [ml](<https://devfeed.tech/tags/ml.md>), [models](<https://devfeed.tech/tags/models.md>), [prevent](<https://devfeed.tech/tags/prevent.md>)

### AI overview

The article describes an intern project measuring when large language models predict outcomes by recalling facts from pretraining rather than using new information. It discusses cricket match previews as a test dataset, differences between pre-cutoff and post-cutoff predictions, and challenges including changes in the data and inconsistent model output.

### Source excerpt

The following is part of a series of posts about 2026 summer intern projects - for more, see "What the interns have wrought, special jumbo 2026 edition"

## What the TDOE interns have wrought, 2026 edition

DevFeed: [What the TDOE interns have wrought, 2026 edition](<https://devfeed.tech/articles/what-the-tdoe-interns-have-wrought-2026-edition-62849.md>)

Original publisher: [Read original article](<https://blog.janestreet.com/tdoe-intern-projects-2026/>)

Author: James Somers

Published: 2026-09-30T00:00:00Z

Content type: article

Language: en

Sources: [Jane Street](<https://devfeed.tech/sources/jane-street.md>)

Topics: [openjdk](<https://devfeed.tech/topics/openjdk.md>), [decision-making](<https://devfeed.tech/topics/decision-making.md>), [Tool](<https://devfeed.tech/topics/tool.md>)

Tags: [dashboard](<https://devfeed.tech/tags/dashboard.md>), [developers](<https://devfeed.tech/tags/developers.md>), [internship](<https://devfeed.tech/tags/internship.md>), [ocaml](<https://devfeed.tech/tags/ocaml.md>), [outage](<https://devfeed.tech/tags/outage.md>)

### AI overview

Jane Street describes its 2026 Trading Desk Operations Engineer intern projects. The work included investigating outages, writing OCaml code to automate error-prone processes, building dashboards with traders, and improving trading workflows and systems. The article also explains how interns ramp up in complex trading domains and begins a discussion of reconciling voice trades.

### Source excerpt

The following is part of a series of posts about 2026 summer intern projects--for more, see "What the interns have wrought, special jumbo 2026 edition"

## Can you use autoregressive diffusion to generate market data?

DevFeed: [Can you use autoregressive diffusion to generate market data?](<https://devfeed.tech/articles/can-you-use-autoregressive-diffusion-to-generate-market-data-62847.md>)

Original publisher: [Read original article](<https://blog.janestreet.com/can-you-use-autoregressive-diffusion-to-generate-market-data/>)

Author: James Somers

Published: 2026-09-30T00:00:00Z

Content type: article

Language: en

Sources: [Jane Street](<https://devfeed.tech/sources/jane-street.md>)

Topics: [llm-reasoning](<https://devfeed.tech/topics/llm-reasoning.md>), [Machine learning](<https://devfeed.tech/topics/machine-learning.md>)

Tags: [autoregressive](<https://devfeed.tech/tags/autoregressive.md>), [diffusion](<https://devfeed.tech/tags/diffusion.md>), [flow-matching](<https://devfeed.tech/tags/flow-matching.md>), [internship](<https://devfeed.tech/tags/internship.md>), [quantitative-finance](<https://devfeed.tech/tags/quantitative-finance.md>)

### AI overview

A summer research intern built an event-level generative model of market data using autoregressive diffusion. The post says DDPM diverged while flow matching performed much better, and that fully continuous diffusion did not fit the jaggedness of real market data.

### Source excerpt

The following is part of a series of posts about 2026 summer intern projects - for more, see "What the interns have wrought, special jumbo 2026 edition"

## What the interns have wrought, special jumbo 2026 edition

DevFeed: [What the interns have wrought, special jumbo 2026 edition](<https://devfeed.tech/articles/what-the-interns-have-wrought-special-jumbo-2026-edition-62852.md>)

Original publisher: [Read original article](<https://blog.janestreet.com/wrought-2026/>)

Author: James Somers

Published: 2026-09-30T00:00:00Z

Content type: article

Language: en

Sources: [Jane Street](<https://devfeed.tech/sources/jane-street.md>)

Topics: [post-training](<https://devfeed.tech/topics/post-training.md>), [llm-reasoning](<https://devfeed.tech/topics/llm-reasoning.md>), [NUMA](<https://devfeed.tech/topics/numa.md>)

Tags: [apprenticeship](<https://devfeed.tech/tags/apprenticeship.md>), [generative](<https://devfeed.tech/tags/generative.md>), [intern](<https://devfeed.tech/tags/intern.md>), [internship](<https://devfeed.tech/tags/internship.md>), [llms](<https://devfeed.tech/tags/llms.md>)

### AI overview

Jane Street reviews selected projects built by its 2026 summer interns across ML research, Linux engineering, IT, strategy and product, and trading desk operations. The examples include a generative model of market data and experiments on reducing memorization in LLMs.

### Source excerpt

It's time again to review some of what Jane Street's interns built over the summer, though this year we're going to try to cover more projects and in more depth. In particular we're including projects from teams and roles whose work has historically been hard to talk about publicly--like ML research and trading desk operations. There's just been too much exciting work in these areas to leave them out.

## A study of sequence weighting at scale

DevFeed: [A study of sequence weighting at scale](<https://devfeed.tech/articles/a-study-of-sequence-weighting-at-scale-20145.md>)

Original publisher: [Read original article](<https://blog.janestreet.com/a-study-of-sequence-weighting-at-scale/>)

Author: Alex Renda

Published: 2026-09-14T00:00:00Z

Content type: article

Language: en

Sources: [Jane Street](<https://devfeed.tech/sources/jane-street.md>)

Topics: [scaling laws](<https://devfeed.tech/topics/scaling-laws.md>), [dataset](<https://devfeed.tech/topics/dataset.md>), [Deep neural networks](<https://devfeed.tech/topics/deep-neural-networks.md>), [Chinchilla scaling law](<https://devfeed.tech/topics/chinchilla-scaling-law.md>)

Tags: [data](<https://devfeed.tech/tags/data.md>), [machine-learning](<https://devfeed.tech/tags/machine-learning.md>), [model](<https://devfeed.tech/tags/model.md>), [scaling-laws](<https://devfeed.tech/tags/scaling-laws.md>), [training](<https://devfeed.tech/tags/training.md>)

### AI overview

The article studies how sequence weighting affects language-model training across model scales. It reports non-monotonic behavior: small-to-medium models increasingly learn data-specific patterns in proportion to sequence weights, while large models can learn all patterns in the data more independently of those weights.

### Source excerpt

TL;DR: We study the scaling laws of data weighting across in-house and open-weight LMs, finding non-monotonic behavior across scales. We vary the weight assigned to sequences during training and measure how strongly the model's loss reduction on a sequence depends on the sequence's weight. Taken together, our results are consistent with a general trend: as models transition from small to medium scale, they transition from learning general patterns independent of data weight to learning data-specific patterns proportional to the data weights. As models then transition from medium to large scale they are able to learn all patterns present in the data, once again independent of data weight.

## Can you design a chip? Announcing the protocol emulator ASIC competition

DevFeed: [Can you design a chip? Announcing the protocol emulator ASIC competition](<https://devfeed.tech/articles/can-you-design-a-chip-announcing-the-protocol-emulator-asic-competition-20207.md>)

Original publisher: [Read original article](<https://blog.janestreet.com/protocol-emulator-asic-competition/>)

Author: Benjamin Devlin

Published: 2026-09-10T00:00:00Z

Content type: release

Language: en

Sources: [Jane Street](<https://devfeed.tech/sources/jane-street.md>)

Topics: [Chip design](<https://devfeed.tech/topics/chip-design.md>), [Open Source](<https://devfeed.tech/topics/open-source.md>), [Reverse Engineering](<https://devfeed.tech/topics/reverse-engineering.md>)

Tags: [ai](<https://devfeed.tech/tags/ai.md>), [ai-assisted](<https://devfeed.tech/tags/ai-assisted.md>), [architecture](<https://devfeed.tech/tags/architecture.md>), [asic](<https://devfeed.tech/tags/asic.md>), [can](<https://devfeed.tech/tags/can.md>), [can-bus](<https://devfeed.tech/tags/can-bus.md>), [challenge](<https://devfeed.tech/tags/challenge.md>), [chip](<https://devfeed.tech/tags/chip.md>), [chip-design](<https://devfeed.tech/tags/chip-design.md>), [cpu](<https://devfeed.tech/tags/cpu.md>), [cycles](<https://devfeed.tech/tags/cycles.md>), [debugging](<https://devfeed.tech/tags/debugging.md>), [design](<https://devfeed.tech/tags/design.md>), [emulator](<https://devfeed.tech/tags/emulator.md>), [engineering](<https://devfeed.tech/tags/engineering.md>), [ethernet](<https://devfeed.tech/tags/ethernet.md>), [firmware](<https://devfeed.tech/tags/firmware.md>), [layout](<https://devfeed.tech/tags/layout.md>), [protocol](<https://devfeed.tech/tags/protocol.md>), [verification](<https://devfeed.tech/tags/verification.md>)

### AI overview

The challenge is to design an open-source, general-purpose protocol emulator ASIC: a small CPU that reads and writes pins and counts cycles so firmware can implement protocols with precise timing. Unlike a die with fixed UART, SPI, and I2C blocks, the design should be reprogrammable after fabrication to support new protocols within its timing and I/O limits. UART, SPI, and I2C are starting points; low-speed USB and 10Mbit Ethernet are stretch goals, with JTAG, SWD, PS/2, and CAN bus also suggested. FPGA testing of RTL is encouraged for those with access. The organizers welcome varied design languages and verification methods, including formal methods, constrained-random tests, and AI-assisted verification. Favorite designs will be fabricated and provided on dev boards. The excerpt identifies IHP's 130nm CMOS5L process through Tiny Tapeout, but cuts off before the rest of the rules.

### Source excerpt

Last month, we asked you to reverse engineer a chip from nothing but its layout and teased a bigger challenge. Results and our favorite writeups are coming soon. In the meantime, here's our next challenge! This time, you're designing the chip, and we'll pay to fabricate our favorite designs! We're particularly interested in projects with unique functionality, as well as those that demonstrate novel approaches to design and verification methodologies!

## Can you reverse engineer an ASIC?

DevFeed: [Can you reverse engineer an ASIC?](<https://devfeed.tech/articles/can-you-reverse-engineer-an-asic-20156.md>)

Original publisher: [Read original article](<https://blog.janestreet.com/can-you-reverse-engineer-an-asic/>)

Author: Anish Singhani

Published: 2026-08-05T00:00:00Z

Content type: article

Language: en

Sources: [Jane Street](<https://devfeed.tech/sources/jane-street.md>)

Topics: [Reverse Engineering](<https://devfeed.tech/topics/reverse-engineering.md>), [Chip design](<https://devfeed.tech/topics/chip-design.md>), [Neural Network](<https://devfeed.tech/topics/neural-network.md>)

Tags: [asic](<https://devfeed.tech/tags/asic.md>), [layout](<https://devfeed.tech/tags/layout.md>), [puzzle](<https://devfeed.tech/tags/puzzle.md>), [reverse-engineering](<https://devfeed.tech/tags/reverse-engineering.md>)

### AI overview

The puzzle provides an ASIC's unlabeled final layout, sample inputs, and outputs, and asks solvers to recover its netlist, identify its function, then find the string that makes the output signal go high. The text explains the path from Verilog through synthesis and physical placement and routing to a GDS file used for fabrication. It suggests inspecting the layout for functional clues, initially ignoring a section that generates but does not affect the output, and simulating the circuit; solvers should toggle before each input attempt. The supplied excerpt does not include the puzzle materials or its solution.

### Source excerpt

Earlier this year we published a puzzle that handed you a complete neural network and asked you to figure out what it did. The response was great, so we've made another one! This time, we're going much deeper down the tech stack. For this puzzle we've designed a chip, but we're only giving you the layout. A crash course in how chips get made Modern chips start life as code.

## Using OxCaml to implement type-safe reference counting between OCaml and Python

DevFeed: [Using OxCaml to implement type-safe reference counting between OCaml and Python](<https://devfeed.tech/articles/using-oxcaml-to-implement-type-safe-reference-counting-between-ocaml-and-python-20199.md>)

Original publisher: [Read original article](<https://blog.janestreet.com/oxcaml-typesafe-reference-counting-python/>)

Author: Nicolas Trangez

Published: 2026-06-15T00:00:00Z

Content type: article

Language: en

Sources: [Jane Street](<https://devfeed.tech/sources/jane-street.md>)

Topics: [OCaml](<https://devfeed.tech/topics/ocaml.md>), [typesafety](<https://devfeed.tech/topics/typesafety.md>), [Python](<https://devfeed.tech/topics/python.md>), [atomic reference counting](<https://devfeed.tech/topics/atomic-reference-counting.md>), [Machine learning](<https://devfeed.tech/topics/machine-learning.md>)

Tags: [data-analysis](<https://devfeed.tech/tags/data-analysis.md>), [data-types](<https://devfeed.tech/tags/data-types.md>), [high-performance](<https://devfeed.tech/tags/high-performance.md>), [machine-learning](<https://devfeed.tech/tags/machine-learning.md>), [memory](<https://devfeed.tech/tags/memory.md>), [memory-management](<https://devfeed.tech/tags/memory-management.md>), [ocaml](<https://devfeed.tech/tags/ocaml.md>), [parallelism](<https://devfeed.tech/tags/parallelism.md>), [python](<https://devfeed.tech/tags/python.md>), [reference](<https://devfeed.tech/tags/reference.md>), [type-safe](<https://devfeed.tech/tags/type-safe.md>)

### AI overview

Jane Street's PyOCaml bridge can delay Python object deallocation when Python objects are represented as OCaml values, which is problematic for large data frames or scarce resources such as GPU memory. The excerpt says OxCaml extensions let library authors encode prompt deallocation in a typesafe way, allowing the compiler to rule out Python programs with these promptness problems. The approach uses explicit reference counting between Python and OCaml and draws on borrowing; the excerpt begins explaining Python's reference-counted object lifecycle but cuts off before detailing the implementation.

### Source excerpt

Jane Street is known for being an OCaml shop, but for years now Python has been our second major programming language, acting as the primary tool for data analysis and (especially importantly these days) machine learning. Most of our traders and researchers think and write in Python, even as the majority of our infrastructure is written in OCaml.

## Formal methods and the future of programming

DevFeed: [Formal methods and the future of programming](<https://devfeed.tech/articles/formal-methods-and-the-future-of-programming-20167.md>)

Original publisher: [Read original article](<https://blog.janestreet.com/formal-methods-at-jane-street-index/>)

Author: Yaron Minsky

Published: 2026-06-07T00:00:00Z

Content type: opinion

Language: en

Sources: [Jane Street](<https://devfeed.tech/sources/jane-street.md>)

Topics: [Formal verification](<https://devfeed.tech/topics/formal-verification.md>), [Programming](<https://devfeed.tech/topics/programming.md>), [typesafety](<https://devfeed.tech/topics/typesafety.md>)

Tags: [agentic-coding](<https://devfeed.tech/tags/agentic-coding.md>), [agents](<https://devfeed.tech/tags/agents.md>), [building-software](<https://devfeed.tech/tags/building-software.md>), [c](<https://devfeed.tech/tags/c.md>), [code](<https://devfeed.tech/tags/code.md>), [formal-methods](<https://devfeed.tech/tags/formal-methods.md>), [hardware](<https://devfeed.tech/tags/hardware.md>), [models](<https://devfeed.tech/tags/models.md>), [proofs](<https://devfeed.tech/tags/proofs.md>), [software](<https://devfeed.tech/tags/software.md>), [specifications](<https://devfeed.tech/tags/specifications.md>), [synthesis](<https://devfeed.tech/tags/synthesis.md>), [tools](<https://devfeed.tech/tags/tools.md>), [type-systems](<https://devfeed.tech/tags/type-systems.md>)

### AI overview

Jane Street says agentic coding has shifted its view of formal methods: models can make verification tools easier for more people to use, changing the cost-benefit calculation. The excerpt contrasts this possibility with seL4, whose verification reportedly took 25 person-years for 8,700 lines of C, and notes that formal methods had previously seemed too expensive for most software, even critical systems. The passage begins to discuss growing benefits as models write more code, but cuts off before explaining them.

### Source excerpt

I've been telling people for the last 25 years that Jane Street as an organization was just not interested in formal methods.

## strace-ui, Bonsai\_term, and the TUI renaissance

DevFeed: [strace-ui, Bonsai\_term, and the TUI renaissance](<https://devfeed.tech/articles/strace-ui-bonsai-term-and-the-tui-renaissance-20213.md>)

Original publisher: [Read original article](<https://blog.janestreet.com/strace-ui-bonsai-term-and-the-tui-renaissance/>)

Author: James Somers

Published: 2026-05-26T00:00:00Z

Content type: article

Language: en

Sources: [Jane Street](<https://devfeed.tech/sources/jane-street.md>)

Topics: [Text-based user interface](<https://devfeed.tech/topics/tui.md>), [debugging](<https://devfeed.tech/topics/debugging.md>), [debug tools](<https://devfeed.tech/topics/debug-tools.md>), [Bonsai](<https://devfeed.tech/topics/bonsai-rx.md>)

Tags: [but](<https://devfeed.tech/tags/but.md>), [can](<https://devfeed.tech/tags/can.md>), [debugging](<https://devfeed.tech/tags/debugging.md>), [exploring](<https://devfeed.tech/tags/exploring.md>), [filter](<https://devfeed.tech/tags/filter.md>), [flag](<https://devfeed.tech/tags/flag.md>), [follow](<https://devfeed.tech/tags/follow.md>), [found](<https://devfeed.tech/tags/found.md>), [it](<https://devfeed.tech/tags/it.md>), [it-s](<https://devfeed.tech/tags/it-s.md>), [make](<https://devfeed.tech/tags/make.md>), [this](<https://devfeed.tech/tags/this.md>), [threads](<https://devfeed.tech/tags/threads.md>), [tool](<https://devfeed.tech/tags/tool.md>), [trace](<https://devfeed.tech/tags/trace.md>), [ui](<https://devfeed.tech/tags/ui.md>), [useful](<https://devfeed.tech/tags/useful.md>), [want](<https://devfeed.tech/tags/want.md>), [with](<https://devfeed.tech/tags/with.md>), [work](<https://devfeed.tech/tags/work.md>)

### AI overview

strace-ui makes strace traces easier to scan and explore interactively. It assigns short labels to processes and threads, formats structs, and displays buffers as hexdumps. Users can hide syscalls, jump between calls using the same file descriptor, filter by FD, PID, or excluded PID, and open a syscall's man page. It attempts to track FD reuse and follow descriptors across forks, but the excerpt says this works poorly when attached to a process that already has open FDs. DNS resolution can display hostnames alongside ports. The excerpt then introduces Bonsai, an OCaml framework for composing interactive UIs from purely functional state machines, as part of the context for strace-ui's development; it cuts off before explaining further.

### Source excerpt

We've always found strace useful but somewhat hard to work with. Its output is often inscrutable, it's hard to follow subprocesses or threads, and if you want to filter syscalls you have to rerun the trace with a flag for each one. What you want in debugging is a tool for exploring, refining, etc., but strace can make this difficult.

## Using group theory to explore the space of positional encodings for attention

DevFeed: [Using group theory to explore the space of positional encodings for attention](<https://devfeed.tech/articles/using-group-theory-to-explore-the-space-of-positional-encodings-for-attention-20221.md>)

Original publisher: [Read original article](<https://blog.janestreet.com/using-group-theory-to-explore-positional-encodings-attention/>)

Author: Alok Puranik

Published: 2026-04-22T00:00:00Z

Content type: article

Language: en

Sources: [Jane Street](<https://devfeed.tech/sources/jane-street.md>)

Topics: [Language models](<https://devfeed.tech/topics/language-models.md>), [Machine learning](<https://devfeed.tech/topics/machine-learning.md>), [Temporal data](<https://devfeed.tech/topics/temporal-data.md>)

Tags: [analysis](<https://devfeed.tech/tags/analysis.md>), [blog-post](<https://devfeed.tech/tags/blog-post.md>), [language-models](<https://devfeed.tech/tags/language-models.md>), [machine-learning](<https://devfeed.tech/tags/machine-learning.md>), [ml](<https://devfeed.tech/tags/ml.md>)

### AI overview

The article examines positional encodings for attention using group theory. It argues that formalizing desirable properties leaves only a few valid families, most of which are already used in real systems, while also identifying a technically valid but apparently unexplored class.

### Source excerpt

Attention is a computational primitive at the core of modern language models, allowing internal representations to reference and influence each other. It's how these models handle sequential data in the first place.

## Can you reverse engineer our neural network?

DevFeed: [Can you reverse engineer our neural network?](<https://devfeed.tech/articles/can-you-reverse-engineer-our-neural-network-20157.md>)

Original publisher: [Read original article](<https://blog.janestreet.com/can-you-reverse-engineer-our-neural-network/>)

Author: Ricson Cheng

Published: 2026-02-24T00:00:00Z

Content type: article

Language: en

Sources: [Jane Street](<https://devfeed.tech/sources/jane-street.md>)

Topics: [Neural Network](<https://devfeed.tech/topics/neural-network.md>), [Reverse Engineering](<https://devfeed.tech/topics/reverse-engineering.md>), [ctf](<https://devfeed.tech/topics/ctf.md>)

Tags: [mechanistic-interpretability](<https://devfeed.tech/tags/mechanistic-interpretability.md>), [ml](<https://devfeed.tech/tags/ml.md>), [neural](<https://devfeed.tech/tags/neural.md>), [neural-network](<https://devfeed.tech/tags/neural-network.md>), [process](<https://devfeed.tech/tags/process.md>), [puzzle](<https://devfeed.tech/tags/puzzle.md>), [research](<https://devfeed.tech/tags/research.md>)

### AI overview

The puzzle gives solvers the full specification and weights of a neural network and asks them to reverse engineer its behavior using mechanistic interpretability. The network outputs 0 for almost all inputs and was designed to frustrate brute force approaches such as backpropagating a nonzero output to the input, so solving it requires reasoning about what the network is doing. The post says it will trace a solver's process, but the supplied excerpt stops before that account or the solution.

### Source excerpt

A lot of "capture-the-flag" style ML puzzles give you a black box neural net, and your job is to figure out what it does. When we were thinking of creating our own ML puzzle early last year, we wanted to do something a little different. We thought it'd be neat to give users a complete specification of the neural net, weights and all. They would then be forced to use the tools of mechanistic interpretability to reverse engineer the network--which is a situation we sometimes find ourselves facing in our own research, when trying to interpret features of complex models.

## Results from the Advent of FPGA Challenge

DevFeed: [Results from the Advent of FPGA Challenge](<https://devfeed.tech/articles/results-from-the-advent-of-fpga-challenge-20148.md>)

Original publisher: [Read original article](<https://blog.janestreet.com/advent-of-fpga-challenge-2025-results/>)

Author: Anish Singhani

Published: 2026-02-11T00:00:00Z

Content type: news

Language: en

Sources: [Jane Street](<https://devfeed.tech/sources/jane-street.md>)

Topics: [Advent of Code](<https://devfeed.tech/topics/advent-of-code.md>)

Tags: [advent-of-code](<https://devfeed.tech/tags/advent-of-code.md>), [challenge](<https://devfeed.tech/tags/challenge.md>), [community](<https://devfeed.tech/tags/community.md>), [competition](<https://devfeed.tech/tags/competition.md>), [fpga](<https://devfeed.tech/tags/fpga.md>), [hardware](<https://devfeed.tech/tags/hardware.md>), [results](<https://devfeed.tech/tags/results.md>)

### AI overview

The Advent of FPGA Challenge drew 213 submissions from 46 countries; 59% of participants were students and 32% industry professionals. Hardcaml was tried in at least one problem by 151 submissions, alongside established HDLs and a broad range of other languages. Days 1 and 3 often used greedy state-machine approaches, while Days 4 and 7 suited shift-register techniques; graph problems on Days 8 and 11 required clever hardware translations. Judges valued detailed, reusable write-ups and ambitious solutions, such as demonstrating results on hardware or using unusual backends. The excerpt begins describing retired Canadian engineer Eric Pearson, who solved both parts of all 12 days in SystemVerilog across ASIC, Altera, and Xilinx flows, but cuts off before giving further details.

### Source excerpt

At the end of last year, we decided to try something new: a challenge that would run alongside Advent of Code, where we asked the community to show us how they could design hardware to solve the same problems. We had no idea what level of participation to expect, but we received a huge number of submissions, many of which were incredibly creative!

## I design with Claude more than Figma now

DevFeed: [I design with Claude more than Figma now](<https://devfeed.tech/articles/i-design-with-claude-more-than-figma-now-20180.md>)

Original publisher: [Read original article](<https://blog.janestreet.com/i-design-with-claude-code-more-than-figma-now-index/>)

Author: Edwin Morris

Published: 2026-02-05T00:00:00Z

Content type: opinion

Language: en

Sources: [Jane Street](<https://devfeed.tech/sources/jane-street.md>)

Topics: [AI-enhanced product development](<https://devfeed.tech/topics/ai-enhanced-product-development.md>), [Claude](<https://devfeed.tech/topics/claude.md>), [UI/UX Design](<https://devfeed.tech/topics/ui-ux-design.md>), [User Experience](<https://devfeed.tech/topics/user-experience.md>), [software-development](<https://devfeed.tech/topics/software-development.md>)

Tags: [ai](<https://devfeed.tech/tags/ai.md>), [claude](<https://devfeed.tech/tags/claude.md>), [design](<https://devfeed.tech/tags/design.md>), [docs](<https://devfeed.tech/tags/docs.md>), [figma](<https://devfeed.tech/tags/figma.md>), [iteration](<https://devfeed.tech/tags/iteration.md>), [keyboard-shortcuts](<https://devfeed.tech/tags/keyboard-shortcuts.md>), [llms](<https://devfeed.tech/tags/llms.md>), [ocaml](<https://devfeed.tech/tags/ocaml.md>), [prototype](<https://devfeed.tech/tags/prototype.md>), [pull-request](<https://devfeed.tech/tags/pull-request.md>), [sql](<https://devfeed.tech/tags/sql.md>)

### AI overview

The author's design workflow at Jane Street shifted from specs, Figma mockups, and implementation reviews toward building prototypes directly in the codebase with Claude. They describe writing a problem and proposal, prompting Claude in an editor, getting basic functionality working, iterating, testing in a development environment with users, and submitting the finished feature. In a JSQL prompting prototype, Claude enabled repeated refinements to controls, shortcuts, copy, prompts, and confirmation messages; the author says this improved the actual artifact and avoided days of design and engineering back-and-forth. Earlier attempts with Copilot, Cursor, and Gemini had disappointed them, while AI support became useful for unfamiliar technologies and hands-on design work. The excerpt ends as the author begins describing how the workflow developed, so it does not provide the full account.

### Source excerpt

For a long time I was skeptical of LLMs--whenever I reached for them I was disappointed by the results. Last year I tried Copilot and Cursor to tweak a game I'd built, and neither generated working changes. At a previous job I tried Gemini to outline product briefs and generate wireframes, but ended up throwing them all away. Every time I tried LLMs it was for something I was already good at, and they did a worse job than I would have.

[Next page](<https://devfeed.tech/sources/jane-street.md?cursor=WyIyMDI2LTAyLTA1VDAwOjAwOjAwKzAwOjAwIiwgIjEwZmUwODMyLWY3MWQtNGNmMS1iY2M0LTU2YjFmZDAxZmQ4MyJd>)