Skip to content
View mrapaul's full-sized avatar

Block or report mrapaul

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
mrapaul/README.md

Arindam Paul — trading infrastructure, exchange connectivity, AI for regulated operations

Engineering leader. Twenty years building systems where failure is expensive — sub-millisecond trading infrastructure, life-critical clinical workflows, and the metadata layer behind a product Apple shipped.

Track record

90+ trading venues Exchange connectivity at IMC Financial Markets — FIX, MDP 3.0, iLink, ITCH, Pillar
Apple Music Classical Built the 5M-recording classical metadata database at Primephonic, acquired by Apple in 2021
2 US patents Emergency radiology workflow — closed-loop critical notification and escalation chains
630+ securities Goldman Sachs reconciliation platform, delivered in 4 months
3 months ETF market-making desk, zero to production

Currently at ING, building FX swap market data systems for capital markets. Previously Flow Traders, Primephonic, IMC Financial Markets, Lakeland Healthcare, Nighthawk Radiology, and Rockwell Automation. Mentor at Cambridge.

Building now

nasdaq-itch-ouch — building a Nasdaq market driver in Java across nine parts: a working FIX session on QuickFIX/J, then SoupBinTCP, OUCH, ITCH over MoldUDP64, and a zero-allocation ring-buffer pipeline. Every part ships code that runs against a local venue simulator and a number that was measured rather than asserted.

jvm-latency-receipts — four reproducible ways JVM latency measurements mislead, with the full percentile distributions committed. A 200ms stall reported as a 5.8µs p99.9 by a naive harness and 202,244µs by a corrected one — same run, same stall, ~35,000× apart. Measuring correctly changes the answer by more than most of the things people benchmark.

CodeSpark Tech — AI products for work that can't fail: clinical quality, medical language, and human-risk governance for regulated operations.

dutch-asr-benchmarks — public evaluation receipts for Vlotspraak, our Dutch speech-recognition engine. 3.5% WER on Common Voice NL under the community benchmark's own NIST sclite tooling, against 4.3% for the previous best on the board. Ships with a same-model control run and a train-on-test contamination check, because a benchmark number without a receipt isn't a result.

The stack

Languages — Java, Kotlin, Python, C++

Low latency — LMAX Disruptor ring buffers, lock-free data structures, DPDK kernel bypass, NUMA-aware memory, zero-allocation hot paths, JVM and GC tuning

Platform — Spring Boot, Kubernetes, Docker, AWS, GCP, PostgreSQL, MongoDB

Domains — market data, order management, reconciliation, and exchange protocol integration across equities, options, ETFs, and FX

Working together

Leadership — CTO, VP Engineering, and Head of Technology roles in capital markets and regulated AI.

Advisory and specialist delivery — architecture review, three-to-six-month buildouts, embedded technical leadership.

Amsterdam, globally mobile. Engagement details at arindampaul.com/engage.

Writing

Most recent: Why I Mentor at Cambridge (And What the Next Generation Teaches Me)

All posts at arindampaul.com/blog.


arindampaul.com · LinkedIn · arindam.paul@me.com

Pinned Loading

  1. MoveMoney MoveMoney Public

    Account-to-account transfer service API in Java — concurrent request handling with transaction integrity guarantees.

    Java

  2. jvm-latency-receipts jvm-latency-receipts Public

    Four reproducible ways JVM latency measurements mislead — coordinated omission, warm-up, means, and hot-path allocation — with committed HdrHistogram receipts.

    Java

  3. nasdaq-itch-ouch nasdaq-itch-ouch Public

    Building a Nasdaq exchange driver in Java — from a QuickFIX/J session to a zero-allocation OUCH/ITCH pipeline. Nine parts, each with running code and measured numbers.

    Java