Software Engineer · Agentic AI Systems

Systems that reason,
not just execute.

I'm Kshitiz Raj, a software engineer at Barclays. I build full-stack platforms and event-driven backends by day, and multi-agent, reasoning-first systems on the side — the kind that plan, delegate, and validate rather than run a fixed script.

IExperience

Two years, one bank, a shift from forecasting risk to running infrastructure.

Both roles at Barclays, in Pune — first building the ML that flags project risk, now building the systems that run in production.

Mar 2025 — Current

Software Engineer, BA4

Barclays · Pune, IN

Own delivery across CI/CD, GraphQL APIs, and reporting infrastructure for systems running in production on Red Hat OpenShift.

  • ~10 reposMigrated CI/CD pipelines to AWS using CloudFormation and Service Catalog, moving the team off Bitbucket Stash onto GitLab with reusable pipeline components.
  • GraphQLBuilt and maintained Node.js GraphQL APIs for data aggregation, giving client teams flexible, on-demand access to shared data.
  • JWTImplemented authentication and authorization with JWT tokens, enforcing access control and session integrity across application layers.
  • SQL ServerDesigned and optimized complex queries and stored procedures for high-volume reporting workflows.
  • OpenShiftDeployed and operated UI and API services across multiple Red Hat OpenShift clusters for scalability and consistency.

Jul 2024 — Feb 2025

Technical Graduate Developer, BA3

Barclays · Pune, IN

Built the ML systems that forecast project health, taking a manual RAG-status process into a model teams could trust.

  • 82% → 96%Built and deployed XGBoost and LightGBM models to forecast project RAG (Red/Amber/Green) status, lifting accuracy from 82% to 96% at 91% precision — cutting false-positive risk alerts.
  • End-to-endRan full data science workflows — EDA, feature engineering, hypothesis testing, statistical validation — across historical, time-series, and categorical data to isolate real risk indicators.
  • 4h → 1.5hCut ML training time 63% through efficient feature pipelines, model tuning, and stratified k-fold cross-validation.
IIWhere I Build

Five layers of the same stack, from reasoning down to reliability.

I don't think of AI work and infrastructure work as separate disciplines — the same instinct for clear boundaries and predictable failure applies to both.

Agentic AI Systems

01

Multi-agent architectures with a centralized root agent for planning, delegation, and session-level context — RAG retrieval paired with explicit validation stages, not just chained prompts.

Google ADKLangChainLangGraphA2ARAG

Backend & Distributed Systems

02

Event-driven services that hold up under real load — Kafka pipelines, async task execution, and APIs built to be consumed by more than one kind of client.

Node.jsFastAPIGraphQLKafkaWebSockets

Data & Applied ML

03

Models built to be trusted in production, not just accurate in a notebook — feature pipelines, validation, and tuning aimed at cutting false positives, not chasing a leaderboard score.

XGBoostLightGBMFeature EngineeringTF-IDF

Cloud & Reliability

04

Infrastructure that a team can operate, not just Claude can deploy once — CI/CD as reusable components, multi-cluster deployments, consistent environments.

AWSRed Hat OpenShiftCloudFormationGitLab CI/CD

Frontend

05

Interfaces that stay out of the way — fast, mobile-first, and built to match the reliability of the backend behind them.

React.jsNext.jsTailwind CSS
IIIWork

What I build when nobody's asking me to.

Side projects that push into territory the day job doesn't — governed multi-agent systems, event-driven platforms, and a few things built purely out of curiosity.

IVProcess

I build systems the way I'd want to inherit them.

Whether it's an agent pipeline or a reporting service, the sequence is the same — because the failure modes are the same.

1

Map the flow

Before writing code, trace how data and requests actually move through the system — not the version on the architecture diagram.

2

Design the pipeline

Draw clear boundaries between reasoning, execution, and context — so the system stays auditable as it grows, not just fast to demo.

3

Ship and harden

Deploy with CI/CD, monitor what's running, and iterate against real usage — reliability is a feature, not an afterthought.

VSkills

The toolkit, laid flat.

Languages

TypeScriptJavaScriptPythonC / C++

Applied AI

Google ADKAWS BedrockAWS AgentCoreLangChainLangGraphA2AReAct Tool UseRAGVector DatabasesPrompt EngineeringMCP-style Context Propagation

Backend & Systems

FastAPINode.jsExpress.jsREST APIsJWTOAuth 2.0Event-Driven ArchitectureWebSocketsKafka

Data & Storage

PostgreSQLMySQLMongoDBRedisPrisma ORMSupabase

Cloud & DevOps

AWSDockerRed Hat OpenShiftCI/CD (GitLab)ELK Stack

Frontend

React.jsNext.jsViteTailwind CSSHTML / CSS

ML & NLP

XGBoostLightGBMFeature EngineeringTF-IDF
VIEducation

Foundations.

2020 — 2024

SRM Institute of Science & Technology

B.Tech, Computer Science & Engineering

Chennai, TN · CGPA: 9.05 / 10

2017 — 2020

Delhi Public School

AISSCE (12th) · AISSE (10th)

Bokaro Steel City, JH · 90.20% · 88.4%

Certifications & Courses

  • Barclays — Agentic AI Digital Credential
  • Barclays — Gen AI Digital Credential
  • Google — Build Intelligent Agents with ADK
  • Google — Agents Intensive Capstone Project
  • 100xDevs — Full Stack Development (0–1 & 1–100)
  • NPTEL — Data Science for Engineers
  • Red Hat — Enterprise Linux Fundamentals

VII · Let's Talk

Building something that needs to reason, not just run?

Open to full-stack and agentic-AI engineering conversations. The fastest way to reach me is email.

LinkedInGitHubXkshitizraj.com
© 2026 Kshitiz RajSoftware Engineer · Pune, IN