Software Engineer — Backend & Platform Systems

Hi, I'm Mahendra Singh Khati

I build backend systems for production: APIs, async workflows, and platforms designed for reliability, scale, and clear ownership.

View Case Studies Get in Touch

About

Production backend, end to end

Software engineer with 6+ years building backend systems for SaaS and service environments. I take work from design through rollout and keep it stable in production.

Recent focus: APIs, event-driven workflows, messaging, caching, observability, and AI-assisted development, with reliability and recoverability treated as core requirements.

Years experience

50%

Throughput improvement

35%

Latency reduction

Career

Work experience

Six years across SaaS and service environments, from API delivery and performance work to architecture spikes, production debugging, and multi-engineer backend execution.

Pune, Maharashtra · Aug 2024 — Present

Software Engineer · Entrata

Event-driven architecture AI workflows Platform migration

Own backend initiatives across asynchronous workflows, scheduled processing, platform modernization, and production stabilization in a SaaS environment.

Architected asynchronous workflows using RabbitMQ and AWS SQS for decoupled background processing and reliable message-driven systems.
Built EventBridge-based scheduled and event-triggered workflows for recurring and conditional task execution.
Implemented idempotent message consumers with structured retry and failure handling for consistency and fault tolerance.
Integrated Redis for caching and rate control, reducing database pressure and improving API response times.
Designed push notification workflows for Android and iOS with reliable backend coordination.
Contributed to PHP 8.3 platform migration, dependency cleanup, and production script stabilization.
Migrated PSI Logger from file-based logging to remote logging to improve observability and reduce server I/O overhead.
Leveraged CloudWatch and New Relic for monitoring, log analysis, and proactive performance issue detection.
Led architecture spikes, authored TRDs, drove POCs, and coordinated multi-engineer backend execution for major epics.
Contributed to AI maintenance workflows using Amazon Bedrock, LiteLLM, RAG, and queue orchestration.

Prabhadevi, Maharashtra · Oct 2021 — Aug 2024

Backend Engineer · WeAssemble

50% throughput gain 35% lower DB latency

Designed and owned RESTful API development using Laravel and FastAPI for critical application workflows and data-intensive processes.

Increased web scraping throughput by 50% using thread pool-based concurrency optimization.
Reduced database latency by 35% through indexing strategy redesign and complex query refactoring.
Built structured ETL pipelines to clean, normalize, and validate large-scale scraped datasets before ingestion.
Diagnosed and resolved legacy date-time inconsistencies affecting multiple dependent application flows.
Collaborated with backend and frontend teams to align API contracts and reduce integration rework.
Participated in architectural discussions, code reviews, and Agile ceremonies.

Andheri, Maharashtra · Feb 2020 — Sep 2021

Backend Developer · Sanjar E Solutions

API foundations Secure services

Developed scalable RESTful APIs and backend services using Laravel, Node.js, and MySQL.

Designed normalized database schemas and optimized queries for efficient data retrieval and performance.
Implemented authentication and authorization mechanisms to secure backend services.
Collaborated with frontend developers and QA teams to ensure stable production deployments.
Documented technical specifications, architectural decisions, and deployment workflows.

Education

Academic background

Jun 2016 — May 2019 · Vasai, India

B.Sc. in Information Technology

St. Gonsalo Garcia College

Jun 2014 — Feb 2016 · Vasai, India

Higher Secondary Certificate (H.S.C)

New English School And Junior College

Technical Profile

Skills & expertise

Hands-on backend experience across REST APIs, event-driven architecture, distributed messaging, cloud delivery, observability, and AI-assisted development with production guardrails.

Core technical stack

Runtime

Languages

PHP Python JavaScript

Frameworks

Laravel FastAPI Node.js

Data & messaging

Databases

MySQL PostgreSQL MongoDB

Messaging & events

RabbitMQ AWS SQS AWS EventBridge

Caching

Redis

Platform & observability

Cloud & infrastructure

AWS S3 CloudWatch Terraform Docker CI/CD

Observability

New Relic CloudWatch Amplitude

AI systems

Intelligent systems

Amazon Bedrock LiteLLM RAG AI agents LLM applications

AI-assisted development

ChatGPT Gemini Claude Cursor IDE

Architecture & backend concepts

REST APIs and event-driven architecture
Idempotency, retry mechanisms, and concurrency control
Performance optimization and legacy refactoring
System design, TRDs, POCs, and production debugging

Delivery & engineering practice

Git, shell scripting, and CI/CD workflows
Linux-based backend operations and deployment support
Cross-team API contract alignment and code reviews
Structured documentation for maintainable releases

Measured outcomes

50% Throughput improvement

35% Latency reduction

Event-driven workflows with RabbitMQ, SQS, and EventBridge
Production-safe PHP 8.3 migration and platform modernization

Things I Built

Selected case studies

Choose a case study below to read the full problem, constraint, architecture, impact, and stack.

Event-Driven Workflow Platform

Private · Entrata

Problem

Critical operational alerts depended on synchronous backend execution paths and tightly coupled retry logic, making delayed retries, repeated escalations, and long-running workflows difficult to manage reliably under failure conditions.

Constraint

The system required configurable recurring notifications with strict idempotency guarantees, resilient retry behavior, and fault isolation between workflow stages. Failures in downstream processing could not block upstream application flows or create duplicate notification delivery.

Architecture

Designed and implemented an event-driven workflow orchestration platform using RabbitMQ, AWS EventBridge, and Amazon SQS to support delayed execution, recurring processing, and distributed retry coordination across asynchronous services.

Built queue-backed orchestration pipelines capable of continuously scheduling and re-queuing events at fixed intervals until acknowledgment conditions or expiration windows were satisfied. Introduced asynchronous execution boundaries between processing stages to isolate failures, improve recovery behavior, and reduce operational pressure on synchronous application paths.

Integrated multi-channel notification delivery pipelines using Twilio, FCM, and APNs to support scalable SMS and mobile push notification workflows across distributed backend services. Implemented deterministic retry handling, idempotent processing patterns, and explicit failure recovery mechanisms to maintain consistency under partial failure scenarios.

Impact

Improved reliability of long-running asynchronous workflows
Eliminated blocking retry logic from synchronous application flows
Reduced duplicate event processing through idempotent workflow execution
Increased operational resiliency through distributed failure isolation
Established scalable event-driven processing patterns across backend services

PHP RabbitMQ AWS SQS AWS EventBridge Twilio FCM APNs PostgreSQL

Smart Maintenance Assistant

Private · Entrata

Problem

Operational support workflows relied on repetitive form-based intake processes and manual triage, creating delays in issue resolution, fragmented context gathering, and inefficient routing of operational tasks.

Constraint

The platform needed to leverage AI for conversational intake, classification, and recommendation generation without allowing non-deterministic model output to directly impact critical backend workflows. All downstream orchestration had to remain observable, recoverable, and auditable.

Architecture

Built an AI-powered conversational operations platform using Amazon Bedrock, LiteLLM, and retrieval-based architectures to transform structured intake workflows into adaptive, context-aware interactions.

Designed retrieval pipelines that analyzed incoming queries against historical operational data, prior resolution patterns, and internal knowledge sources to provide guided troubleshooting, dynamically request additional contextual information, and reduce unnecessary escalation into backend systems.

Implemented asynchronous orchestration pipelines using AWS SQS to transition unresolved interactions into structured backend workflows enriched with conversational metadata and operational context. Combined LLM-driven reasoning with deterministic queue-backed processing to ensure backend execution remained fault-tolerant, observable, and operationally safe.

Introduced modular AI orchestration layers to separate conversational reasoning, retrieval logic, workflow execution, and downstream system integration, improving maintainability and scalability across the platform.

Impact

Reduced manual operational triage through AI-assisted conversational workflows
Improved contextual data collection during intake and escalation flows
Increased operational efficiency by minimizing unnecessary backend escalations
Established reliable AI-to-workflow orchestration boundaries for safe production execution
Built scalable retrieval and orchestration pipelines for future AI-driven operational automation

Amazon Bedrock LiteLLM RAG Python AWS SQS PostgreSQL

Distributed Data Intelligence Platform

Private · WeAssemble

Problem

Data-intensive ingestion and analytics workflows processing external datasets were bottlenecked by synchronous scraping pipelines, inefficient database access patterns, and inconsistent legacy data behavior, resulting in slow reporting generation and reduced downstream performance.

Constraint

The platform needed to significantly increase ingestion throughput while preserving data quality, supporting concurrent processing, and maintaining predictable ETL behavior across continuously changing external data sources.

Architecture

Designed and implemented a distributed data ingestion and analytics platform using concurrent scraping and asynchronous processing patterns. Built scalable scraping and automation pipelines using Python and Selenium to extract structured and semi-structured data from dynamic web sources.

Developed ETL workflows using Pandas for large-scale cleaning, normalization, enrichment, and aggregation before persisting optimized analytical datasets into PostgreSQL. Introduced query optimization and indexing strategies to improve read-heavy analytical workloads and reduce reporting latency.

Built high-performance API layers using FastAPI to expose processed datasets and analytical insights while supporting scalable concurrent access patterns. Generated visualization and reporting pipelines using Matplotlib for downstream analytics consumption.

Impact

Improved scraping throughput by 50%
Reduced database and reporting latency by 35%
Increased reliability of large-scale ingestion workflows
Established scalable ETL pipelines for predictable analytics processing

Laravel FastAPI PostgreSQL Python Selenium Pandas Matplotlib Concurrency Query optimization

Contact

Let's connect

Open to conversations about backend architecture, distributed systems, event-driven platforms, and production reliability work.