(00) Index

Software Engineer — Mumbai, India

Building systemsthat scale. Engineering experiencesthat matter.

Software Engineer focused on backend systems, full-stack engineering, cloud infrastructure and production AI.

Scroll

(01)Engineering identity

I build software thatsolves real problems —and make complex systems simpler.

I'm a software engineer at ICICI Lombard in Mumbai, working across the stack on insurance products customers use every day. Some weeks that means migrating a legacy Angular journey to Next.js and moving its business rules into .NET services. Other weeks it means building the MCP tools that let an AI assistant look up a policy — safely, behind role-based access.

I care about the parts users never see but always feel: payload sizes, API round-trips, access boundaries and recovery plans. I'd rather measure a problem than argue about it, and I'd rather ship one reusable piece than write the same code twice.

  1. 01

    Build for real users

    Customer journeys first; architecture serves the person on the other side of the screen.

  2. 02

    Simplify complexity

    Move orchestration to where it belongs so every layer does less.

  3. 03

    Measure performance

    Lighthouse, payload sizes, token counts — numbers before opinions.

  4. 04

    Design for reliability

    Health checks, access boundaries, and recovery as configuration.

  5. 05

    Automate the repetitive

    Reusable widgets, tools and services instead of the same code twice.

  6. 06

    Keep learning

    From Angular to Next.js, OpenAI to Vertex AI — the stack changes; the habit doesn't.

(02)Selected work

Production systems,explained.

Six pieces of production work at ICICI Lombard — AI, product, modernization, integrations and reliability. Every number shown carries its measurement scope.

  1. 01 / 06

    Production GenAI · RAG · MCP · AI orchestration

    KYP — AI-Powered Insurance Assistant

    Customers and teams needed accurate answers about policies, workflows and products — grounded in enterprise systems, not guessed by a model.

    My role — Built the MCP tool layer · contributed to retrieval and integration

    12,000+

    Peak daily active users

    Reported peak DAU for the production assistant

    4

    MCP tools built

    Policy lookup, workflow guidance, company profile, product RAG

    • Python
    • FastAPI
    • MCP
    • RAG
    • Vertex AI
    • Vertex AI Search
    • Next.js
    Read case study: KYP — AI-Powered Insurance Assistant
    KYP architecture: a user query is routed by a FastAPI service through role-based access control to one of four MCP tools, whose context grounds a Vertex AI response.INPUTUser queryNEXT.JS CLIENTFastAPI serviceCODE-BASED ROUTINGRBAC · SIGNED TOKEN CLAIMSMCP TOOLSPolicy lookupCORE POLICY SYSTEMWorkflow guidanceSTEP-BY-STEPCompany profileANSWERSProduct RAGVERTEX AI SEARCHVertex AIGROUNDED GENERATIONGrounded response12,000+ PEAK DAU · REPORTED
    Simplified, public-level view of the documented architecture
  2. 02 / 06

    Full-stack product engineering · Microfrontends · Performance

    RYD Raksha — Motor Insurance Product

    A new digital motor-insurance product had to ship a complete customer journey — and move very large reference datasets to the browser without punishing users.

    My role — Owned frontend delivery · technical guidance to the backend team

    17 MB → 6 MB

    API payload

    Vehicle make-model & RTO reference data, Brotli + .NET output caching (approx.)

    6 MB → 1.1 MB

    Browser cache

    LZ-String compression, 1-day LocalStorage TTL (approx.)

    • Next.js
    • React
    • TypeScript
    • Zustand
    • .NET 10
    • Vite
    • Webpack
    Read case study: RYD Raksha — Motor Insurance Product
    RYD Raksha: reference-data API payload compressed from about 17 MB to 6 MB with Brotli and output caching, then the browser copy from about 6 MB to 1.1 MB with LZ-String; a KYC microfrontend is embedded in other journeys.REFERENCE DATA · MAKE-MODEL + RTO① API PAYLOAD — .NET OUTPUT CACHE (ECS MEMORY)17 MB≈6 MB · BROTLI② BROWSER COPY — LOCALSTORAGE, 1-DAY TTL6 MB≈1.1 MB · LZ-STRINGREUSABLE KYC MICROFRONTEND · SCRIPT + CSSKYC widgetVITE / WEBPACK BUNDLERYD Raksha journeyMotor renewalPost-payment buying
    Simplified, public-level view of the documented architecture
  3. 03 / 06

    Enterprise modernization · Next.js · .NET · Performance

    Motor Renewal Modernization

    Legacy Angular renewal journeys carried business logic in the browser and orchestrated many services client-side — slow to load and hard to change.

    My role — Primary hands-on engineer · guided interns

    37 → 93

    Lighthouse score

    On the migrated renewal journey

    ~27%

    Faster page load

    On the migrated renewal journey

    • Angular
    • Next.js
    • React
    • TypeScript
    • .NET
    • ASP.NET Core
    • EF Core
    Read case study: Motor Renewal Modernization
    Motor renewal modernization: before, an Angular client called several services itself; after, a Next.js client calls one frontend-ready .NET API that talks to the core policy system directly. Lighthouse rose from 37 to 93 on the migrated journey.BEFOREAngular clientBUSINESS RULES IN UISERVICE 1SERVICE 2SERVICE 3SERVICE 4IntermediaryXML CORE INTEGRATIONAFTERNext.js clientSSR / SSGASP.NET Core APIFRONTEND-READY RESPONSESCore policy systemJSON (V2.0) SERVICESPostgreSQLMIGRATED FROM ORACLE93LIGHTHOUSE · FROM 37~14% FEWER API CALLS~27% FASTER LOADMIGRATED RENEWAL JOURNEY
    Simplified, public-level view of the documented architecture
  4. 04 / 06

    Generative AI · Streaming · Personalization

    Travel Itinerary GenAI Sales Accelerator

    Turn a traveller's intent into a personalized itinerary — and connect each activity to the travel-insurance benefits that matter for it.

    My role — Built the generation workflow and streaming

    ~65%

    Less token/memory use

    Through context/data optimization — not a latency figure

    • Python
    • FastAPI
    • LLM
    • Semantic retrieval
    • Google Maps
    • NDJSON streaming
    • Next.js
    Read case study: Travel Itinerary GenAI Sales Accelerator
    Travel itinerary GenAI: user intent goes through context selection and generation with validation, and results stream to the client as NDJSON lines that render progressively.Traveller intentNEXT.JSContext selectionSEMANTIC RETRIEVALGenerate + validateFASTAPI · FALLBACKS~65% LESS TOKEN / MEMORY USEGoogle MapsNDJSON STREAM · ONE OBJECT PER LINE (ILLUSTRATIVE){"type":"day","n":1}{"type":"activity",...}{"type":"benefit",...}{"type":"place",...}{"type":"day","n":2}CLIENT RENDERS AS LINES ARRIVE
    Simplified, public-level view of the documented architecture
  5. 05 / 06

    Backend engineering · Event-driven workflows · Integrations

    TripSecure+ / Voyager Travel Insurance

    Flight-disruption claims depend on external flight data, customer documents and multi-step forms that users don't finish in one sitting.

    My role — Built backend services and integrations (Engineer Trainee)

    • .NET
    • ASP.NET Core
    • AWS S3
    • Cirium APIs
    • REST APIs
    • Webhooks
    Read case study: TripSecure+ / Voyager Travel Insurance
    TripSecure+: Cirium flight data and webhook events feed ASP.NET Core services that apply database-driven claim rules, store documents in S3 and persist partial submissions.INBOUNDCirium flight dataDELAY · CANCELLATIONWebhook eventsEVENT-DRIVENCustomer submissionMULTI-STEPASP.NET Core servicesCLAIM WORKFLOWClaim rulesDATABASE-DRIVENDocumentsAWS S3Partial stateRESUMABLE→ FLIGHT-DISRUPTION CLAIM PAYOUTSIMPLIFIED, PUBLIC-LEVEL VIEW
    Simplified, public-level view of the documented architecture
  6. 06 / 06

    Cloud infrastructure · AWS ECS · Resilience

    Cross-Region Disaster Recovery & Reliability

    If a region fails, services must come back elsewhere — without someone writing code during the incident.

    My role — Contributor

    • AWS ECS
    • Docker
    • Health checks
    • Monitoring
    • Configuration as code
    Read case study: Cross-Region Disaster Recovery & Reliability
    Cross-region disaster recovery: a domain endpoint points at ECS services in the primary region; configuration managed in code repoints it to the recovery region during DR, while health checks monitor services in both.Domain endpointCUSTOMER TRAFFICDR CONFIG · MANAGED IN CODEPRIMARY REGIONECS service 1ECS service 2ECS service 3RECOVERY REGIONECS service 1ECS service 2ECS service 3● HEALTH CHECKS · NO RTO/RPO FIGURES CLAIMED
    Simplified, public-level view of the documented architecture

(03)Technical capabilities

Breadth across the stack,depth where it shipped.

No skill bars. Each group points to the production work that used it.

(04)Experience

From traineeto end-to-end ownership.

  1. Aug 2025 — Present

    Mumbai, India

    Associate Software Engineer

    ICICI Lombard General Insurance

    From delivering individual features to owning journeys end to end — frontend, backend APIs, core-system integration, production GenAI tooling and DR readiness.

    • Primary hands-on engineer on the Angular-to-Next.js migration of motor renewal and two-wheeler journeys; Lighthouse 37 → 93 on the migrated renewal journey.
    • Designed frontend-ready ASP.NET Core APIs that moved business rules and orchestration out of the client (~14% fewer API calls on the renewal journey).
    • Owned frontend delivery of RYD Raksha and built a reusable KYC microfrontend used by other journeys.
    • Built the four MCP tools, code-based routing and RBAC behind KYP, a GenAI assistant with a reported peak of 12,000+ daily active users.
    • Contributed to live Oracle-to-PostgreSQL migration, core-system JSON (v2.0) services replacing XML, and config-driven cross-region DR on AWS ECS.
  2. Feb 2025 — Aug 2025

    Mumbai, India

    Engineer Trainee

    ICICI Lombard General Insurance

    Backend services and integrations in C# / ASP.NET Core.

    • Built ASP.NET Core services and webhook-driven travel-insurance workflows: Cirium flight data, database-driven claim rules, S3 document storage and partial-submission state.
    • Contributed to the backend for commercial-vehicle one-click renewal on the core policy system, productionized in ~2 weeks.
    ProjectsTripSecure+

(05)Measured impact

Numbers withtheir context attached.

Reported figures from specific projects and journeys — not totals, and not continuously live.

(06)Education, awards & leadership

The foundationbehind the work.

Education · 2021–2025

B.E. Information Technology

Thakur College of Engineering and Technology

9.83/10

CGPA
Department Topper

  • Award

    HighFlier Award

    ICICI Lombard

    2026

  • Award

    Best Research Paper Award

    Multicon-W

    2024

  • Leadership

    Technical Head

    ACM TCET

    College

(06.1) Academic projects

College work, kept separate from professional experience.

  • Diabetes Decoded

    Final-year project using a hybrid CNN-LSTM model. The associated research paper received the Best Research Paper Award at Multicon-W 2024.

    • CNN-LSTM
    • Python
    • Deep learning
  • VANI — Voice Assistance for Navigation & Interaction

    Voice-driven navigation assistant with Whisper-based speech transcription.

    • JavaScript
    • Leaflet
    • Flask
    • Whisper

(06.2) Curious about

AI and emerging technology/Software architecture/Developer tools/Robotics and electronics/Building useful products/Gaming and science fiction

(07)Contact

Have a complexproblem?Let's engineera better solution.