19 Years · Systems Architect · Constraint Modeling · Graph Thinking
The Through-Line
I start every problem by making it visible. A diagram, a mind map, a system sketch. Clarity precedes good architecture.
For nineteen years I've done the same thing across payments infrastructure, search systems, and commerce platforms: stand outside a complex system, feel where the pressure is accumulating, and make one small move that reorganizes everything.
I think in layers: what does the customer need today, what does the system need to become, and what's the smallest step that moves toward both.
I don't chase trends. I build foundations.
One Shape, Many Bodies
Only the present holds the real form
The past is written, the future is read. The present alone is real — and every hard problem lives at the crossing, where abstraction becomes actual.
Every system I've built has this shape. The four below are four crossings I was hired to hold.
Signature Work
Workflow Orchestration for Commerce
Meta · 2020–2024
Multiple product teams couldn't onboard. One major product line couldn't integrate at all. Diagnosed the root cause as a structural impossibility in the platform's type system. Not a bug — a category-level design error. Spent two years making the case. Executed a 1,000-file migration with zero incidents. Built the replacement: a DAG-based workflow orchestration engine with a novel type system, version control, and visualization tooling. Five verticals, fifteen-plus products, billions of requests per year, multi-billion dollar payment volume. Grew to a team of six. Platform became so critical a dedicated team was created to assume ownership.
Zero-incident migration·Billions of requests/year·Platform outlived the team 5 verticals·15+ products
AWS Data Exchange — Search Vision & Launch
AWS · 2017–2020
Designed and launched the Marketplace feature of AWS Data Exchange, contributing to the service's path to GA. Built the search ingestion pipeline that decoupled indexing from publishing. Created the search vision mind map that aligned all ADX leadership on product direction — became the team's architectural reference document. Go-to architect; could diagram the entire ADX architecture on demand.
Aligned leadership·Enabled GA·Became the team's architectural reference
EFD Project Rescue — CashPro Global Payments
Bank of America · 2016–2017
Took over a failing payments delivery multiple sprints in. Diagnosed root problems, realigned the team, and delivered with zero defects within two months. Enabled major client migrations off legacy systems. Built automation reducing two-week processes to two to three days. Led four agile teams building payment processing for enterprise banking clients.
Meta's largest advertisers needed finer control over conversion values before events entered the pipeline. I built the upstream rule-evaluation layer that gave them that control — and ran the delivery as a case study in AI-operated development: Claude wrote the code, reviewed it, and responded to reviewer feedback, grounded in my orchestration platform's codebase. Design doc to production in two weeks. The build system transferred to six engineers. Unblocked alpha partners representing $400M+. The finding: code generation is solved — the open problem is the engineer operating the system.
$400M+ unblocked·AI-operated SDLC, hands off·Doc → production in 2 weeks
Independent Research
Working with AI carries a hidden tax: every session starts cold, and the person with the most context pays the most — re-explaining what they already know. The Rapport Problem is my research into that tax and its inversion. Instead of rebuilding rapport every session, you write the contract that lets a cold mind — human or AI — begin faithfully. Intent is declared up front, acceptance is held apart from production, and the engineer surfaces only at genuine decisions.
The mechanism is a write-store / read-store split. Work runs through two sealed branches: a Producer builds the artifact, an Auditor builds the acceptance test, and each verifies the other's output: the Producer checks that the criteria are sound, the Auditor checks that the artifact meets them. Only the Auditor seals. They iterate until a boundary decision is reached, and every pass — failure or success — is appended to a canonical store that is write-only and never queried directly.
Read-only views are derived from that store only after decision, each built for a specific purpose: one store, many views. A cold mind reads a settled view rather than the live stream, so it begins from a declared contract instead of an empty context. Its own decisions then become the next writes.
Personal work·Oct 2025 – present·Self-funded, full time·Publishing August 2026
Full Record
2025 – 26Ads Pixel Foundation — Meta2020 – 24Workflow Orchestration for Commerce — Meta2018 – 20Search Integration, Data Exchange — AWS, New York2017 – 18Search Foundation, Marketplace — AWS, Seattle2016 – 17EFD Payment-File Orchestrator — Bank of America2015 – 16CashPro Last Mile — Bank of America2012 – 15Canonical Reporting Engine — Bank of America2010 – 11Styling Generator, "How to Think in CSS" — Bank of America2009 – 17Payment Initiation Engine — Bank of America2007 – 09AAA + BCCR — Infosys / BofA, Pune
What People Say
"His super power is his ability to distill his big product ideas into documents / artifacts that others can understand."
Peer, AWS Data Exchange
"Getting lighting in meeting rooms just right. Rajan has quietly delivered a number of components that were stretch goals in our launch, and he did it with very little overhead."
General Manager, AWS Data Exchange
"He is not just a cornerstone but a foundation on which the entire team can rely upon."
SVP, Senior Technology Manager, Bank of America
"You have consistently shown an ability to solve difficult technical problems with clean and well-articulated solutions which are above the standard at your level."