Shridhar Pathak

Exit

CAREER RECORD SYSTEM v0.9
LOADING SUBJECT: PATHAK, SHRIDHAR
! WARNING: RECORD CONTAINS TIMESTAMPS BEYOND SYSTEM EPOCH
! DISPLAYING ANYWAY

> ORIGINS.TXT

+----------------------------------------------------+
|  ACADEMIC RECORD                          [SEALED] |
+----------------------------------------------------+
|  KVPY FELLOWSHIP ............... 2005-2009         |
|    Young Scientist, Engineering Stream             |
|    National science scholarship, India             |
|                                                    |
|  B.E. COMPUTER SCIENCE ......... JUL 2009          |
|    Pune Institute of Computer Technology           |
|                                                    |
|  RESEARCH FELLOW, TIFR ......... 2009-2010         |
|    GRAPES-3 / CERN                                 |
|    Scintillator monitoring algorithms              |
|    ROOT framework, C++                             |
|                                                    |
|  M.S. COMPUTER SCIENCE ......... MAY 2012          |
|    University of Texas at Dallas                   |
|    Distributed Systems / ML / Algorithms           |
|                                                    |
|  M.B.A., MIT SLOAN ............. EXP MAY 2026      |
|    Cross-registered, Harvard Business School       |
+----------------------------------------------------+
          

The subject begins in physics computing. Detector data from a cosmic ray experiment, written in C++, read on a screen that shows one colour. Nothing about this suggests a career in conversational machines. That is generally how it goes.

Everything after this point is, from where you are standing, in the future. The terminal does not mind.

> FULL RECORD, ABBREVIATED

  • 2012-2015 AMZN.REGISTER Point of sale and payments. Fraud detection. Java, C++. Seattle and Bangalore.
  • 2015-2018 AMZN.RETAIL Southeast Asia launch. Three year timeline reduced to nine months. 26 teams aligned.
  • 2018-2023 AMZN.ALEXA Machines that answer when spoken to. Ten countries. 40 direct reports.
  • 2023-.... MODERNA Language models applied to medicine. 20+ applications. 35+ engineers.
  • 2025-.... [STEALTH] Record incomplete. Subject is building something and has not said what.

END OF FILE. PRESS A YEAR TO CONTINUE.

‹ Career Register Edit

Payments, in the cloud

Software Development Engineer, Amazon.com · June 2012 to June 2015

Amazon Register

He joins Amazon the month after finishing his master's, and lands on the point of sale team. Amazon Register is the company's move into physical retail payments, which in 2012 is where everything interesting is happening.

The work is distributed services and fraud detection: global merchant onboarding, financial portal integrations, and payment flows that cannot fall over. Java and C++. He is tech lead across Seattle and Bangalore, which means the working day has no obvious end.

2

Continents, one codebase

1000s

Merchants onboarded

24/7

Fault tolerance required

The whole story so far

Written as 2012 would have written it, when "the cloud" still needed the quotation marks.

Live frame · Senior Director, AI & ML

Moderna

July 2023 to present · Cambridge, MA

20+

Enterprise applications on the GenAI roadmap

35+

Person engineering org, built from zero

~30%

Targeted reduction in R&D and clinical timelines

2

Multi-million-dollar partnerships: OpenAI, AWS

Production LLM workflows on AWS for global manufacturing, built to hold up in a regulated environment. The hard part is rarely the model. It is governance, evaluation, and convincing an executive team to fund something whose output is probabilistic.

The 30% figure is a portfolio-level target attached to a roadmap, not a banked result. Worth saying plainly, since this frame is the present tense and the present tense is where people round up.

Full trace

Archival retrieval · advisory instance

On the subject

What follows was assembled from a record its subject maintained himself, in the years when people still wrote their own.

The shape of it

He spent the first half of his working life making systems large and the second half making them accountable. Payments, then logistics, then voice, then medicine. Each move looked lateral at the time and only resolves into a line from here.

What he was known for

Compressing a three year country launch into nine months. Building the first framework that let a voice assistant hand a conversation from one agent to another, which sounded like a novelty and turned out to be the beginning of something. Standing up generative AI inside a pharmaceutical company at a moment when very few regulated industries were willing to try.

The venture

In 2025 he began building the company that eventually became part of how enterprises learned to trust machines. At the time it had three design partners and no revenue, and he described it in public only as an agent evaluation platform. The argument he was making, that you should not deploy an autonomous system you cannot inspect, was not yet obvious. It reads now as a truism, which is the usual fate of a correct early position.

What the record does not contain

The reasoning. Career records preserve outcomes and discard the deliberation that produced them, which is why this instance can tell you what he decided and not why. He is understood to have considered that a reasonable trade.

Index

This instance holds no opinions. It has been asked to say that he would have preferred you email him. shripathmit@gmail.com