Skip to content
Loading, 0 percent
Abhiraj Ghosh

Loading — 000%

AG
Chandigarh, India / UTC+5:30Open to opportunities

Backend Software Engineer

AbhirajGhosh.

Backend engineer, three years in. I mostly work in NestJS and Postgres, and more of that work now involves AI that people actually use, like an assistant I built for the Indian Army that runs completely offline.

Resume
EmailGitHub LinkedIn Linktree
Scroll
01

About

I like the parts of a product most people never see: how the data is laid out, what happens in the background, and who is allowed to do what.

At Talentelgia I led a team of four on a CRM and ERP that 75+ people use every day. I've also spent a lot of time on Indian Army projects: an inventory assistant that works with no internet at all, and an NFC tracking system we rolled out to 14 depots. I'm a Software Engineer L2 there, based in Mohali.

Location
Chandigarh, India — UTC+5:30
Role
Software Engineer L2, Talentelgia
Focus
Backend Systems · Applied AI
Open to
Remote (India) · Relocation to Gurgaon / Noida / Delhi NCR
Education
MCA, Chandigarh University (2024– 26) — CGPA 8.8 · BCA, Chandigarh University (2021–24) — CGPA 7.3
160+
Postgres tables, split across schemas
75+
People using the CRM + ERP daily
14
Army depots running our system
40%
More accurate than baseline RAG

Stack

Backend
NestJSNestJSNode.jsNode.jsTypeScriptTypeScriptBullMQRedisRedisREST APIsExpressExpress.jsSocket.ioSocket.IOMicroservices
Databases
PostgreSQLPostgreSQLPrismaPrismapgvectorMongoDBMongoDBMySQLMySQLSequelizeSequelize
AI / ML
LangGraphLangGraphRAGText2SQLLangChainLangChainOpenAIMistral AIMistralPaddlePaddlePaddleOCR
Cloud & DevOps
AWSDockerDockerGitHub ActionsGitHub ActionsJenkinsJenkinsNGINXNginxAPI GatewayLinuxLinux
Frontend
Next.jsNext.jsReactReact.jsReduxRedux
Tools
GitHubGit / GitHubPostmanPostmanJestJest
02

Experience

Talentelgia Technologies

Feb 2025 — Present
Mohali

Software Engineer L2 — Talentelgia Technologies

  • Led a team of 4 engineers building a CRM and ERP for an education consultancy in NestJS. It has 14 roles that each see only what they should, and 75+ staff use it every day. Access is checked by NestJS guards on JWT auth with refresh-token rotation.
  • Designed the database behind it: 160+ tables in PostgreSQL, with separate schemas for CRM, ERP, payroll and HR. Added indexes and Redis caching in the places where the old system had been slow.
  • Built a two-way HubSpot sync for a US M&A firm. Webhooks go through a BullMQ queue with retries and idempotency keys, so nobody has to clean up duplicate records by hand.
  • Built the backend for an Indian Army inventory assistant. You ask it questions in plain language, and LangGraph agents either search the records or write SQL. That SQL runs as a read-only database user.
  • Gathered requirements on-site with commanding officers, then built and rolled out an NFC tracking system to 14 Army depots. Updating a record used to take hours. Now it takes minutes.
Vaayro Technologies

Nov 2023 — Jan 2025
Mohali

Software Developer — Vaayro Technologies

  • Shipped 10+ features to production across Next.js, Node.js and Express.
  • Added Redis caching that cut page load times by 70%, and reworked real-time tracking on Socket.IO so it ran 40% faster.
  • Mentored 4 junior developers, mostly on debugging, code quality and how to structure a system.
03

Selected Work

Client — Indian Army

Inventory AI Agent

An offline assistant for inventory questions

You ask about inventory in plain language and get answers from the database. LangGraph agents decide whether to search the records (pgvector plus keyword search) or write SQL. The SQL is the risky part, so it runs under a read-only role and can only touch approved tables and query shapes. Paper challans come in through an OCR pipeline built on PaddleOCR. Everything runs on-site on a self-hosted Mistral 8B, with no internet connection at all.

LangGraph / pgvector / PostgreSQL / Text2SQL / PaddleOCR / Mistral 8B

<3 min

to generate a report. It used to take hours

Read the case study Private / Air-gapped deployment

Independent — AI / RAG

MindAgent

Ask questions across your documents

A RAG agent built with LangChain and OpenAI. Search mixes MMR and RRF ranking, which gave better answers than either method alone. It works across several documents, links each answer back to the passages it came from, and supports more than one LLM. It runs on AWS Lambda, set up with CDK, with Pinecone for vectors.

LangChain / OpenAI / Pinecone / AWS Lambda / CDK / TypeScript

40%

more accurate answers than the baseline

GitHubLive

Personal — Distributed systems

Social Media API

A social media backend split into services

Five small services behind one API gateway that handles rate limiting and load balancing, all running in Docker. Redis caches the busiest reads so they don't hit the database every time.

Node.js / Express / Redis / Docker / API Gateway / MongoDB

65%

faster API responses after adding Redis

GitHub

Freelance — Client platform

Mindful by Akansha

A blog for a psychologist

A fast, SEO-friendly blog on Sanity CMS. I set it up so the client can write and publish without needing a developer.

Next.js / Sanity / Tailwind / Vercel

100%

of posts published by the client, no developer needed

Live