Open to conversationsProduct engineering rolesBengaluru / IN

Harshit S / Product Engineer

I build banking systems that hold up when things fail.

Most of my work is around event-driven data, long-running workflows, and APIs that have to keep working under pressure.

40Mrecords searched~3K/slookup traffic90%fewer database reads

Case studiesSee what I've built ExperienceMy resume

Open to conversationsProduct engineering roles

I buildbanking systemsthat hold upwhen things fail.Harshit S / Product EngineerReliable systems for high-stakes workflows.Harshit SProduct Engineer

Distributed Systems / Banking Infrastructure

Most of my work is around event-driven data, durable workflows, high-throughput APIs, and integrations that cannot quietly fail.

See My Work Let's Connect

40MRecords Searched

~3K/sLookup Traffic

90%Fewer Database Reads

8MRecords Validated

02/ ABOUT HARSHIT S

02 / ABOUT HARSHIT S

I like taking the messy, failure-prone parts of banking software and making them boringly reliable.

I’m a Product Engineer at Finacle, working where distributed systems meet high-stakes banking operations. I build infrastructure that moves customer data, coordinates long-running workflows, and keeps external integrations reliable.

Recent work includes a PostgreSQL and Debezium outbox pipeline that replaced nightly transfers, a Redis caching layer serving a gRPC lookup path at roughly 3,000 requests per second, and similarity search across 40 million customer records at 220ms p95.

I took a Temporal prototype from internship experiment to production onboarding, built an idempotent video-KYC webhook framework, and added AWS KMS-backed field encryption for customer PII that helped clear a security audit with zero critical findings.

What I care about is fairly simple: retries should be safe, failures should be visible, tests should resemble production, and the result should be measurable. A system is not finished just because the happy-path demo works.

Application Layer

TypeScriptNestJSReactNext.jsNode.js

Backend & APIs

JavaSpring BootgRPCTemporalSpring BatchIdempotent workflowsWebhooks

Data & Events

PostgreSQLKafkaDebeziumRedispg_trgmGIN indexes

Cloud, Security & Quality

AWS KMSKubernetesDockerJUnit 5EKSHelmTestcontainers

Full-timeAug 2024 - Present

Product Engineer

I work on data pipelines, onboarding workflows, integrations, and performance for enterprise banking products.

01Event-driven data

Replaced nightly transfers with continuous change capture.

Implemented a transactional outbox for a Kafka CDC pipeline, moving customer updates from batch file exchange to near-real-time propagation.

24h to liveupdate latency

PostgreSQL
Kafka
Debezium

02Search at scale

Made duplicate checks fast enough for the request path.

Optimized similarity matching across 40 million customer records using pg_trgm and GIN indexes.

1.8s to 220msp95 per check

PostgreSQL
pg_trgm
GIN

03High-throughput reads

Took repeated customer lookups off the database.

Developed a Redis cache with Kafka-driven invalidation for a gRPC lookup service handling roughly 3,000 requests per second.

90% fewer readsdatabase reads

Redis
Kafka
gRPC

04Durable workflows

Turned an internship prototype into the onboarding backbone.

Migrated customer onboarding to idempotent Temporal workflows with per-step retries, reducing straight-through processing time from six to two minutes.

6m to 2monboarding time

Temporal
Spring Boot
Java

05Vendor integrations

Built one reliable path for video KYC callbacks.

Created a reusable, idempotent Spring Boot webhook framework plus reconciliation for missed callbacks.

3wk to 4dintegration time

Spring Boot
Webhooks
Reconciliation

06Security & platform

Protected PII and made safe test data easier to ship.

Added field-level AES-256 encryption for PAN, mobile, and email with AWS KMS. Also contributed mock vendor APIs to an internal Kubernetes test-data platform adopted by four teams.

0 criticalaudit findings

AWS KMS
NestJS
Docker
EKS
Helm

01 / 06

05 PROJECTS

05 / PROJECTS

A few things I've built.

All work ->

Protocol Engineering2026

MAVLink v2 Telemetry Recorder

An embedded-style C++17 learning project that reconstructs MAVLink v2 frames from a byte stream, decodes selected UAV messages, and records human-readable telemetry to CSV.

AI Systems2024

AI Multi-Agent Orchestration Platform

A production agentic system that decomposes complex tasks, dispatches specialised agents, maintains working memory, and routes tool calls through a typed execution graph.

AI Systems2024

Production RAG Infrastructure for Document Intelligence

End-to-end retrieval-augmented generation pipeline with hybrid dense-sparse retrieval, MMR reranking, and a streaming inference layer. Deployed for a fintech firm with 40K+ document corpus.

Tooling2023

LLM Evaluation Harness

Framework for evaluating language model outputs across factuality, format adherence, and task-specific rubrics. Designed to be extended as models change, not replaced.

AI Systems2023

UPI Transaction Fraud Detection System

Real-time fraud detection pipeline for UPI payment flows, processing 50K+ transactions per minute at sub-100ms decision latency using a tiered rule-based and ML classifier architecture.

01 / 05

06 / THINKING

Notes from figuring things out.

All posts ->

2026-10

A parser that accepts its own packets can still be wrong

What a small MAVLink v2 recorder taught me about state machines, checksums, and the difference between a working demo and wire compatibility.

State machinesChecksumsBinary parsingWire compatibility

07 / Say hello

The easiest way to reach me is on LinkedIn.

If you want to talk about a role, a project, or something I've built, send me a message there. If you would rather look through the code first, you can browse my GitHub.

Want to talk?LinkedInSend me a message Want to see the code?GitHubHave a look around