Domeshwar Kumar

[ 01 / 07 ] About

Senior engineer. Team leader.

I am Domeshwar Kumar — a Senior Backend Engineer & Team Lead with 8+ years of experience architecting production-grade fintech backends, AI-integrated SaaS products, and enterprise payment infrastructure.

At GimBooks (Gim Info Solutions), I designed every critical system from the ground up: a payment engine processing ₹100Cr+ monthly at 99.9% uptime, a Flutter platform with 5M+ downloads and 3M+ active users, real-time CRM dashboards, and 10+ live GST/KYC/banking integrations.

"Focused on clean architecture, AI-driven engineering, and systems built to last. I write code that teams are proud to maintain."

System DesignAI IntegrationPayment SystemsFlutterAPI ArchitectureTeam LeadershipAWS / GCPCI/CD

⚡

Backend Architecture

Django, FastAPI, DRF — APIs serving millions

🤖

AI Engineering

OpenAI, LLMs, intelligent automation

💳

Payment Infra

Multi-gateway, KYC, fraud detection

📊

SaaS at Scale

Zero to 3M+ users ownership

🔄

Async Systems

Celery, Redis, WebSocket pipelines

🐳

DevOps

Docker, Nginx, GitHub Actions, Cloud

[ 03 / 07 ] Experience

Engineering journey

Sept 2020 — Present · 4+ yrs

Gim Info Solutions Pvt. Ltd. (GimBooks)

Full Stack Backend Engineer & Team Lead · Raipur, IN

Led backend engineering for GimBooks — India fintech platform for 3M+ SMBs. Architected payment engine handling ₹100Cr+ monthly at 99.9% uptime with multi-gateway routing, KYC automation (PAN+Aadhaar), real-time fraud detection. Shipped Flutter app from 0 to 5M+ downloads. Built CRM (React+Django+Redis), GST engine, BBPS bill payments from scratch with credit card to vendor and bill payment flows, 10+ live banking integrations. Led engineering hiring, code reviews and cross-team technical leadership.

PythonDjangoDRFFastAPICeleryRedisPostgreSQLFlutterFirebaseAWSGST/KYCRazorpay

Sept 2017 — Sept 2020 · 3 yrs

CodeNicely Software Services LLP

Full Stack Software Engineer · Raipur, IN

Architected scalable ERP systems and built a Document Management System with version control and automated categorization — improving retrieval efficiency by 50%. Delivered client apps across e-commerce, logistics and SaaS using Django, Celery, Django Channels (WebSocket) and PostgreSQL.

PythonDjangoREST APICeleryDjango ChannelsWebSocketPandasPostgreSQL

[ 04 / 07 ] Projects

Systems built at scale

Production systems serving real users. Not side projects.

Production

Payment Infrastructure

GimBooks Payment Engine

▸ Rs.100Cr+/month at 99.9% uptime

Multi-gateway orchestration with KYC gate, fraud detection, idempotent APIs, and credit card to vendor and bill payments.

DjangoCeleryRedisKYC APIs

Problem

Five payment gateways with different APIs, webhook formats, and failure modes had to process crores daily without double-charging or dropping transactions.

My Role

Sole architect of the payment layer; designed every service, review-gated all changes, owned production incidents end to end.

Solution

Provider-agnostic gateway adapter with idempotency keys on every mutation, webhook deduplication via Redis, circuit-breaker failover between gateways, and automated reconciliation jobs in Celery.

Result

Rs.100Cr+ (USD 12M+) processed monthly at 99.9% uptime with zero double-charge incidents since launch.

Live

Bill Payments

BBPS Integration

▸ New revenue line launched from scratch

Bharat Bill Payment System built ground-up: biller discovery, bill fetch, payment and reconciliation, powering credit card to vendor and bill payment flows.

DjangoBBPS APIsCeleryPostgreSQL

Problem

SMBs wanted to pay vendors and bills by credit card, but no in-house rails existed and BBPS has strict flow, retry, and reconciliation requirements.

My Role

Owned the integration end to end: API design, BBPS certification flows, settlement logic, and rollout.

Solution

Built biller discovery, bill fetch, payment execution, and T+1 reconciliation as isolated Django services with Celery retries and an audit trail on every state transition.

Result

Credit card to vendor & bill payments shipped as a new revenue line on top of the existing payment engine.

5M+ Downloads

SaaS Platform

GimBooks SMB App

▸ 0 → 5M+ downloads · 3M+ active SMBs

Mobile-first bookkeeping for Indian SMBs: GSTR filing, e-Way Bill, e-Invoice, in-app payments — backend scaled from zero.

FlutterDjangoGSP APIsFirebase

Problem

A bookkeeping app for non-technical SMB owners had to stay fast and reliable while the user base grew from zero to millions on modest infrastructure.

My Role

Led the backend architecture and the team building it; owned scaling decisions, schema design, and release quality.

Solution

Django/DRF services on AWS with Redis caching on hot paths, Celery for async GST/e-Invoice jobs, and Firebase-backed sync for the Flutter client.

Result

5M+ downloads and 3M+ monthly active businesses with sub-second API responses at peak load.

Live

Real-Time Dashboard

GimBooks CRM

▸ Live ops visibility used daily by leadership

RBAC admin dashboard with live transaction monitoring, Chart.js analytics and Redis-powered real-time notifications.

DjangoReactRedisChart.js

Proprietary fintech — no public demo available.

Live

Compliance Engine

GST Automation Engine

▸ Zero-touch filings for 3M+ businesses

End-to-end GST: GSTR filing, IRN, e-Way Bill, eSign. Provider-agnostic adapter layer with circuit-breaker pattern.

DjangoGSP APIsKYCeSign

Proprietary fintech — no public demo available.

Live

Marketplace API

Vahak Marketplace

▸ APIs serving millions of transporters

Scalable freight marketplace REST APIs with Google recommendation engine integration on AWS.

PythonDjangoMySQLAWS

Proprietary fintech — no public demo available.

AI-Powered

AI SaaS

AI SaaS Automation

▸ Manual back-office effort reduced

LLM-powered workflow automation using OpenAI APIs for intelligent document processing and BI reporting.

PythonFastAPIOpenAICelery

Proprietary fintech — no public demo available.

Open Source

Personal / OSS

AI Experiments Lab

▸ LangChain + OpenAI outside the day job

Personal experiments with LangChain RAG pipelines, OpenAI function calling, and document Q&A agents — the hands-on lab behind the AI Toolkit section.

LangChainOpenAI APIPythonRAG

View GitHub →

[ 05 / 07 ] AI Toolkit

AI tools I use daily

Leveraging AI to engineer faster, smarter and better every day.

🤖

Claude (Anthropic)

AI Assistant

Primary AI coding and architecture assistant. Complex Django system design, code reviews, debugging distributed systems.

Code ReviewArchitecture

💬

ChatGPT (OpenAI)

AI Assistant

Rapid prototyping, explaining complex algorithms, generating boilerplate, exploring alternative engineering approaches.

PrototypingResearch

👾

GitHub Copilot

Code Completion

Real-time completion in VS Code. Accelerates Django views, Celery tasks, DRF serializers and SQL queries.

AutocompleteProductivity

🔗

LangChain

LLM Framework

Production LLM apps: RAG pipelines, document Q&A and intelligent agents with PostgreSQL vector stores.

RAGAgents

⚙️

OpenAI API

LLM Integration

GPT-4, embeddings and function calling integrated into production SaaS for document processing and automation.

GPT-4Embeddings

🤗

Hugging Face

ML Models

Open-source models for text classification and NER for GST document processing in Indian fintech use cases.

Open SourceNER

[ 06 / 07 ] Writing

Notes from production

Short field notes from systems I have built and broken.

Scaling a payment engine to Rs.100Cr+/month

Payments · Reliability

The three things that mattered most: idempotency keys on every mutation, webhook deduplication before business logic ever runs, and circuit-breaker failover between gateways.

Gateways fail in different ways: timeouts that succeeded server-side, duplicate webhooks minutes apart, and partial captures. We treat every incoming webhook as untrusted and replayable — a Redis dedup key rejects repeats before any handler runs. Every payment mutation carries an idempotency key so client retries are safe by construction. When a gateway degrades, a circuit breaker routes new transactions to the next provider and a Celery reconciliation job settles the in-flight ones. Boring machinery, but it is why the volume can grow without the incident count growing with it.

Adding LLM document processing to a Django app

AI/LLM · Django

Function calling plus Celery orchestration turned OpenAI from a demo into a dependable back-office worker — the guardrails matter more than the prompts.

The naive version calls the API inline in a view and prays. The production version pushes every document through a Celery pipeline: extract, classify with a constrained function-calling schema, validate the structured output against hard rules, and only then write to the database — with a human-review queue for anything below a confidence threshold. Cost control came from caching embeddings and batching. The lesson: treat the model like any flaky third-party API — retries, timeouts, validation — and it becomes genuinely dependable.

What BBPS integration taught me about compliance systems

Fintech · Integrations

Compliance rails reward the same design every time: an adapter layer that isolates the provider, retries with audit trails, and reconciliation as a first-class job.

BBPS, GST, e-Invoice, KYC — every compliance integration we shipped converged on the same shape. A provider-agnostic adapter so the business logic never knows which vendor is behind it. Explicit state machines with an audit row for every transition, because when a regulator or a support ticket asks what happened, replaying logs is not an answer. And reconciliation as a scheduled first-class job, not an afterthought — T+1 mismatches are found by us, not by customers. Design for the audit first and the happy path becomes easy.

Senior engineer. Team leader.

Full-stack depth

Engineering journey

Systems built at scale

AI tools I use daily

Notes from production

Let's build something great