Jasmine Chabbra Backend Engineer · Java & Spring Boot
Jasmine Chabbra

Jasmine Chabbra (She/Her)

Associate Software Engineer at Lentra. Java, Spring Boot and REST APIs. Digital lending microservices, 6+ fintech integrations (LOS, CKYC, credit bureau, ACH), SQL, Redis, CI/CD.

Noida, Uttar Pradesh, India· Contact info· GitHub· LinkedIn
Open to work  Software Engineer / SDE, backend. Serving a 1-month notice period.
LN Lentra CU Chitkara University
2.1 yrsBuilding backend at Lentra
6+External financial systems integrated
4Open-source backend services
47Tests across LoanIQ and SentinelPay

About

I'm a backend engineer at Lentra, working on the services behind a digital lending platform. A loan application moves through intake, underwriting and disbursement, and my job is most of the plumbing in between: the Spring Boot services that carry it, and the six-plus external systems they have to talk to on the way.

A lot of that turns out to be failure handling. Bank APIs time out. Callbacks arrive twice, or never. I've spent much of the last year on retry flows that resume at the step that failed instead of starting the whole thing again, and on moving bank-specific logic out of code and into configuration, so onboarding a new bank isn't another branch in a method.

In my own time I build the same kind of system in the open. LoanIQ is a credit underwriting engine where the rule engine makes the call and the model only writes the explanation. SentinelPay scores a payment in milliseconds and leaves the slow thinking to an agent that runs after the payment is already on hold. Both ship with tests, migrations, Docker and CI, because I wanted them to be things a person could clone and actually run.

JavaSpring BootMicroservices REST API designPostgreSQL RedisDockerCI/CD

Experience

Associate Software Engineer

Lentra · Full-time
Jul 2025 – Present · Noida, Uttar Pradesh, India · On-site
  • Design and maintain Java and Spring Boot microservices for digital lending and loan disbursement, delivering REST APIs, business logic, validation and error handling on a production fintech platform.
  • Built and maintained integrations with 6+ external financial systems, including LOS, DMS, CKYC, credit bureau, ACH and decisioning, covering request and response transformation, authentication, callback processing and failure recovery.
  • Architected failure handling and retry flows that reprocess failed operations without repeating completed steps, improving the reliability of financial transactions.
  • Built configurable, lookup-driven data mapping that turns customer data into bank-specific API payloads, removing hardcoded logic for every new bank integration.
JavaSpring BootREST APIs MicroservicesSQLRedis

Full-stack Developer

Lentra · Internship
Oct 2024 – Jun 2025 · Noida, Uttar Pradesh, India · On-site
  • Contributed to Java and Spring Boot backend services and REST APIs supporting financial application and integration workflows.
  • Implemented API validations, data transformations, database operations and integration changes driven by evolving business requirements.
  • Debugged integration issues by tracing application logs, database records, API responses and configuration-driven workflows to root cause.
JavaSpring BootSQLDebugging

Projects

LoanIQ

Oct 2026 – Present

An AI-assisted credit underwriting engine. A regulator can ask two years later why an applicant was declined, and the answer has to come out the same every time it is asked.

Why it is built this way

A model anywhere in the decision path makes the system unauditable, because the same application can score differently on two runs. So the model reads documents and writes the rationale, and nothing else. The decision comes from a deterministic rule engine whose thresholds live in configuration.

  • Five ordered policy rules with hard stops. Every outcome is stored per application, passing rules included, so an audit shows what was considered and not only what failed.
  • Idempotency-Key is a unique column, so a client retry replays the original decision instead of underwriting a second time.
  • A missing bureau score is treated as risk rather than as neutral, which is how thin-file applications get caught instead of waved through.
  • 13 tests covering decision boundaries, determinism and idempotent replay.
Java 21Spring Boot 3.3Spring Data JPA PostgreSQLFlywayRedis TestcontainersDocker
Run the rule engine the real thresholds from application.yml, evaluated in order, stopping where the engine stops
APPROVE 589 / 1000
github.com/JasmineChabbra/LoanIQ ↗

SentinelPay

Oct 2026 – Present

Fraud detection for card and UPI payments, split along the line that decides the architecture: what has to be fast, and what is allowed to be slow.

Why it is built this way

Scoring sits on the payment's critical path, so it stays deterministic, model-free and fixed-cost. Investigation only runs once a payment is already held, which is exactly why it can take seconds and call a model.

  • Five risk signals scored over a single batch of context queries, so adding a signal never adds a query to the hot path.
  • Weights are additive and capped at 100, and no single signal can reach the block threshold alone. One bad signal costs precision instead of declining every customer at once.
  • An agent works held cases through an allow-list of three read-only tools. It recommends, and its recommendation is stored next to the analyst's verdict rather than replacing it, so the two can be compared later.
  • 34 tests across signals, banding boundaries, the agent loop and the full pipeline on a real database.
Java 21Spring Boot 3.3PostgreSQL RedisFlywayTestcontainers DockerGitHub Actions
Score a payment same five signals and weights as the service; account baseline is ₹1,000 average spend with prior history
BLOCK 100 / 100
github.com/JasmineChabbra/SentinelPay ↗

PayRelay

2026

A payout orchestration engine with exactly-once transfers across four bank APIs. A payout that fails halfway has to resume, not restart, so idempotency keys live in both Redis and the database and every step records its own completion.

  • Config-driven integrations for four bank APIs across API key, OAuth2 and HMAC authentication.
  • Step-level resume, retries and a dead-letter queue, so a failed payout never repeats a step that already went through.
  • Zero duplicate payouts under a 25% injected failure rate.
Java 21Spring BootRedis SQLREST
attempt 1 retry validate reserve funds bank transfer confirm timeout skipped skipped bank transfer confirm already recorded complete
The retry does not replay the first two steps. Each step records its own completion against the payout's idempotency key, so a resume starts where it stopped and the money moves once.
github.com/JasmineChabbra/payrelay ↗

PixWipe

Aug 2026 – Sep 2026

A full-stack AI background-removal product, and the one project here where I owned the frontend too: Spring Boot REST APIs over MySQL, a React 18 interface built with Vite and Tailwind, and authentication handled through Clerk instead of rolled by hand.

Spring BootSpring Data JPAMySQL MavenReact 18Tailwind CSS ViteClerk
github.com/JasmineChabbra/PixWipe ↗

Skills

JV

Java and the Spring ecosystem

Two years of production Spring Boot on a live lending platform, plus four services built from scratch in the open.

Java 21 · Spring Boot · Spring Data JPA · Spring Security · Spring MVC · Hibernate · Maven
API

REST APIs and integrations

Integrated 6+ external financial systems, including LOS, DMS, CKYC, credit bureau, ACH and decisioning, across API key, OAuth2 and HMAC authentication.

REST API design · Microservices · Request and response transformation · Callback processing · Postman
REL

Reliability in money flows

Idempotency keys, step-level resume, retries and dead-letter queues, so a failed transaction recovers without repeating a step that already succeeded.

Idempotency · Retry and recovery design · Audit trails · Failure injection
DB

Data

Schema design with versioned migrations, query work against production data, and Redis where a lookup must not hit the database twice.

PostgreSQL · MySQL · SQL · Redis · Flyway
CI

Build and delivery

Every open-source project ships with a multi-stage Dockerfile, Compose for local dependencies, and CI that runs the suite on each push.

Docker · Docker Compose · GitHub Actions · Git · CI/CD
TST

Correctness

47 tests across the two flagship services, covering decision boundaries, determinism, idempotent replay and whole pipelines on a real database.

JUnit 5 · Mockito · Testcontainers · H2

Education

Chitkara University

Bachelor of Engineering, Computer Science and Engineering

Contact info

Email chabbrajasmine275@gmail.com
Location Noida, Uttar Pradesh, India
© 2026 Jasmine Chabbra · Noida, India