Deepak Garg.
There is a category of software engineer whose work almost nobody outside their own team ever sees: the person who builds the framework, the library, the internal service that a hundred other engineers rely on without knowing whose hands wrote it. Deepak Garg has spent much of his career in that category, moving between Bloomberg, PIMCO, and Samsung over roughly two decades, and leaving behind, at each stop, infrastructure that outlasted the projects it was built for.
At Bloomberg LP, where Garg worked as a senior software engineer on the RHUB team from 2015 to 2017, the assignment was regulatory reporting, a domain where the cost of a software defect is not a slow page load but a compliance failure. He helped design RHUB, the firm's regulatory reporting solution, and implemented the MIFID2 version of the platform's service reporting on top of the framework his team had built, at a moment when European regulators were tightening transaction reporting requirements across the industry. Underneath the reporting layer sat a set of engineering problems that had nothing to do with regulation and everything to do with distributed systems: Garg designed a high-throughput RabbitMQ-based demultiplexer and serializer to route messages at scale, built an object-relational mapper on top of Bloomberg's internal aggregate library that supported table joins and inheritance hierarchies, wrote the XML specification for more than one hundred fifty fields spread across multiple databases, and created the service that read counterparty databases and generated the XML data files clients depended on. He also built a library that simplified how updates to data entities were published to the user interface, the kind of small piece of infrastructure that, done well, every other engineer on the team uses without a second thought.
Before Bloomberg, Garg spent two years at PIMCO as a senior software engineer on the FES team, where he worked on the firm's risk management system in coordination with the mortgage analytics group and built a JSON-based workflow orchestration tool for running and managing the firm's analytical models. The tool gave the firm a single format and location for configuring its overnight batch cycles, let batches be linked to one another through defined inputs and outputs, and allowed a failed run to be replayed from its last saved checkpoint rather than restarted from the beginning, a detail that matters enormously to anyone who has had to explain to a trading desk why the previous night's risk numbers are late. That PASS batch system, in its essentials, is still running the firm's workflows today.
Garg returned to PIMCO in 2017 for a five-and-a-half-year stint as lead engineer in Portfolio Analytics, where the framework-building continued at a larger scale. He created an application development framework that handled the basic scaffolding every new service needed: resource management, configuration, credential handling, and data connectors, work that, when done once and done well, saves every engineer who comes after from rebuilding the same plumbing. He built an Avro-based remote procedure call framework, ahead of the firm's move to REST, that handled service description, request and response parsing, dispatch, and error handling. And he built the Executor, a service and client library that let engineers distribute arbitrary Python functions across thousands of workers using four or five lines of code, collapsing what had previously been a distributed-computing problem into something closer to a single function call. A companion project, a wrapper service for Symphony application deployment, cut the lead time for standing up new services by ninety percent.
Garg's earliest engineering work, at Samsung's India operation from 2007 to 2010, looks different from everything that followed but points to the same instinct. He worked on Verve, a scene graph framework written in C++, and wrote a software rendering engine using Google's Skia library, graphics infrastructure a world away from bond pricing and regulatory filings. He holds a bachelor's degree in electronics and communication engineering from the Indian Institute of Technology, Guwahati, and later a master's in computer science from Texas A&M University, a path from hardware-adjacent engineering into the software infrastructure that now underpins some of the financial industry's largest data operations. The throughline across rendering engines, RPC frameworks, and batch orchestration tools is not any single technology but a consistent answer to the same question: rather than solve a problem once, build the thing that lets everyone solve it, indefinitely, after you've moved on.
Author: Faizan Farooqui
