AI ARCHITECTURE & ENGINEERINGPERSONAL PORTFOLIO / 2026

Rahul Yadav.

Applied AI Engineer at Backbase

AI architecture.
Built for real use.

I design and build AI systems for banking: retrieval, API integration, transaction intelligence, and the workflows and evaluation that make them dependable.

Explore my work View resume (PDF) ↗

AI-assisted editorial portrait of Rahul Yadav smiling by a window — THE PERSON BEHIND THE WORK01 / RY

Previously: Intersoft / Bed Bath & Beyond / InfosysSCROLL INTO THE WORK ↓

ARCHITECTURE THROUGH DELIVERY

What I bring to a team.

I work across system design, implementation, integration, and evaluation. My background in test automation shapes how I check AI behavior, handle failures, and make changes reviewable.

ROLES I’M TARGETING

Staff AI Engineer · Forward Deployed Engineer · AI Agent Harness & Evaluation Engineer · AI / Solutions Architect · AI Engineering Lead · AI Engineering Manager

Architecture & applied AI: System design, RAG, agent workflows, LLM evaluation, guardrails, API integration, OpenAPI
AI frameworks & retrieval: LangGraph, LangChain, Azure OpenAI, scikit-learn, LightGBM, Neo4j, Weaviate, FAISS, pgvector
Languages & services: Python, Java, SQL, TypeScript, FastAPI, Django, Celery, PostgreSQL, Redis
Cloud & operations: Azure, Kubernetes, Docker, CI/CD, GitHub Actions, Jenkins, Prometheus, Grafana
Test automation & quality: Selenium WebDriver, REST Assured, Cucumber, TestNG, JUnit, Maven, Postman, regression and API testing

SELECTED ENGINEERING PROJECTS

Systems, with decisions./ 04

01 / BANKING RAG LABAPI DISCOVERY

FROM A QUESTION TO AN OPERATION

Make API operations findable.

A compact retrieval demo that reads OpenAPI specifications, indexes individual operations, and returns ranked matches for natural-language queries.

My focus: Operation-level indexing and a simple, inspectable retrieval baseline.

Result: A portable JSON index and local keyword search over dummy banking specifications. This public sample runs without a model or vector database.

Python · OpenAPI · YAML · JSON · pytest

Explore Banking RAG Lab ↗

Keep the result inspectable.
Show the operation and the words behind the match.

01 / ASK A QUESTION

How do I create a payment?

02 / INSPECT THE MATCH

POST/payments

Create a payment

Operation ID: createPayment
Source: Illustrative OpenAPI spec

WHY THIS MATCH?

The query and operation share the words “create” and “payment”.

SPEC → INDEX → QUERY → MATCH

Authored examples with dummy operations. Buttons switch between preset explanations; no search or banking API runs here.

Inside the retrieval decisionsOPEN THE STORY

THE PROBLEM

An API specification contains many operations. A developer needs a relevant operation without reading the entire document.

THE CHOICE

Index individual operations and start with local keyword search. Keep the result portable and easy to inspect.

THE LIMIT

Keyword matches can miss synonyms and intent. This public sample is a retrieval baseline, separate from enterprise RAG work.

02 / SENTINEL-FIFINANCIAL INTELLIGENCE

FINDING WHAT GETS MISSED

Small leaks.
A bigger picture.

A financial audit system for spotting leakage in small-business transactions and turning findings into reviewable next steps.

My focus: Workflow architecture, backend services, and approval boundaries.

Result: Reviewable audit findings with classification traces and a separate cleanup workflow.

Python · FastAPI · LangGraph · SQLModel · PostgreSQL · Docker

Explore Sentinel-Fi ↗

The important boundary?
Analysis can inform action.
People still approve it.

Sentinel-FiAUDIT NOTE / 001

Where did the
money go?

Recurring paymentsReview ↗

Duplicate subscriptionsInvestigate ↗

Tax categorizationCheck ↗

FINDING → REVIEW → APPROVAL

Illustrated workflow · not a live audit

make the next step clear.

Inside the decisionsOPEN THE STORY

THE PROBLEM

Classification alone does not explain which financial issues deserve attention or what should happen next.

THE CHOICE

Separate the audit workflow from cleanup execution. Keep decision traces and require approval before consequential actions.

THE NEXT QUESTION

How should recovery and repeated requests behave before cleanup is connected to real financial operations?

03 / TRANSACTION AITRY A DESIGN DECISION

Does this really
need an LLM?

One of the most useful AI decisions is knowing when you already have enough information.

Transaction AI combines merchant signals, rules, ML, and selective LLM reasoning. Explore one branch of its fast-mode routing below.

My focus: Classification architecture, routing, and evaluation.

Result: A fast-mode path that skips LLM reasoning when rule and ML signals agree with sufficient confidence.

Python · FastAPI · LightGBM · Next.js · PostgreSQL · Redis

Explore Transaction AI ↗

AT THE ROUTING DECISIONILLUSTRATIVE EXAMPLE

What do the existing signals tell us?

RULESFood & dining92% confidence

MACHINE LEARNINGFood & dining94% confidence

SKIP THE LLM

Enough agreement to take the simpler path.

Both signals agree and meet the 90% fast-mode threshold. This branch does not need an LLM call.

Synthetic signals, not measured model outputs. Illustrates the rule + ML branch with fast mode and LLM routing enabled; no model runs here.

FROM A COMMENT TO A DM

Say the word.
Get the link.

Someone comments a keyword on a post. The bot replies publicly, then sends them the link by DM. Keywords live on a small admin page, so adding one needs no deploy.

Each platform sets its own limits. Threads has no DM API, so there the link goes in the reply itself.

My focus: API integration, configurable workflows, and operational controls.

Result: Keyword-to-reply automation with an admin interface, pacing, and persistent policy pauses.

Python · Meta APIs · Render

Explore the bot ↗

The part that matters most?
Knowing when to slow down.

CommentsPOST / 014

@a.viewer storm 🔥 need this

@creator Thanks for commenting! 🙏 Sent you the link in your DMs ✨

DIRECT MESSAGE · @CREATOR

Hey! 👋 As promised, here’s your link:

example.com/storm

a little different each time.

Illustrated exchange · not a real account

GUARDRAILS, BUILT IN

Hourly caps150 DMs and 200 replies per account, by default
Paced repliesA random 4–9 second gap between comments
Policy pauseA Meta block stops everything for 24 hours
No backlogOlder comments wait for a manual catch-up

More code samples & explorations

API Mapping StudioSchema comparison

BIAN Compliance AuditorAPI conventions

AI Tech NewsroomEditorial workflow · in development

THE EXPERIENCE BEHIND THE WORK

A career built
by building.

I started in software engineering and automation. Today, I design and build AI systems. That journey taught me to care about the architecture and the details that make it hold up.

View resume (PDF)

JUN 2023 — PRESENTCURRENT

Backbase

Applied AI Engineer · Jun 2024 — present

Design and delivery of AI systems for API integration, retrieval, transaction categorization, and test generation. Built retrieval and operation-matching workflows using knowledge graphs, embeddings, and Azure OpenAI; developed and operated Python services with Kubernetes and monitoring.

Python · FastAPI · Azure OpenAI · Neo4j · Weaviate · Kubernetes · Prometheus · Grafana

Previously Senior QA Engineer · Jun 2023 — Jun 2024. Built reusable banking test automation with Java, Selenium, Cucumber, and REST Assured.

JUL 2021 — JUN 2023

Intersoft Data Labs

Senior Test Analyst · Bed Bath & Beyond assignment

Led UI and API test automation, distributed team work, reviewed code, and improved regression and sanity suites with client and engineering teams.

Java · Selenium · REST Assured · Cucumber · Maven · Jenkins

FEB 2017 — JUL 2021

Infosys

Test Analyst

Built Selenium Page Object Model frameworks and automated cross-browser and parallel tests. Worked with users on requirements, API checks, test data, and execution reports.

Java · Selenium · TestNG · Maven · Oracle SQL · Jenkins · Postman

Rahul laughing outdoors beneath a wooden pergola — A little less architecture. A little more me.

BEYOND THE JOB TITLE

There’s a person
behind the projects.

Alongside the engineering, I write Hindi poetry, make cinematic AI videos through madeoftokens, and explain AI through RahulUnpacks.

I like having room for both: the careful thinking that a system needs, and the curiosity that makes you pick up a camera or write a few lines.

RahulUnpacks · AI explainers madeoftokens · Personal AI creations

“Mukammal hu pura, ab koi kami nahi,
Ek goonj si hai mgr shor kahi nhi,”

My words, over at @rahulinks

LET’S MEET

Looking for someone
to take an idea further?

I’d love to hear about your team, the problem you’re solving, and where I could help.

Let’s have a conversation

AI architecture.
Practical engineering.
A person to build with.

Connect on LinkedIn ↗