Abhishek Kumar
Full-Stack & AI Engineer
11 years building end-to-end systems — from NestJS APIs and React frontends to LLM gateways and cloud infrastructure on AWS & Azure.
About
Engineer. Builder.
Occasional trekker.
I'm Abhishek Kumar — a Full-Stack & AI Engineer with 11 years of experience designing, developing, and shipping production-grade systems for businesses that demand quality at every layer.
I don't define myself by the work I've done. I define myself by the work I want to do next. I prefer to keep learning, keep challenging myself, and invest my energy in problems that actually matter.
Whether it's architecting cloud infrastructure across AWS and Azure, integrating LLMs into real-world products, or leading engineering teams to deliver complex platforms — I bring the full stack to every problem. I work to make a better web: fast, accessible, frustration-free, and built to last.
When I'm not pushing code, you'll find me somewhere around Pune — on a trail.
Skills & Stack
Frontend
Backend
Database
Cloud & DevOps
AI & LLM
Security & Tooling
Work
Selected Projects
A mix of products I've built, systems I've architected, and AI infrastructure I've shipped. Real problems, real stakes.
PocketKhata
Accounting Software for Wholesale Businesses
A full-featured accounting platform built for wholesale merchants. PocketKhata lets businesses track bills and payments for both customers and their own company accounts, with rich analytics and account health dashboards — replacing manual ledgers with a clean, reliable digital workflow.
- End-to-end bill & payment tracking per customer and company
- Real-time account stats and balance dashboards
- Deployed on AWS ECS with CloudFront CDN for performance
- Designed for non-technical users — minimal learning curve
LLM Gateway
Production AI Infrastructure with Observability & Guardrails
A self-hosted LLM gateway that routes requests across multiple AI providers, enforces content guardrails, tracks cost and quality via Langfuse, and gives teams a single control plane for all AI interactions — built for production, not experimentation.
- Multi-model routing: OpenAI, Gemini, Claude via LiteLLM
- Full observability with Langfuse — traces, cost, latency
- Input/output guardrails and content moderation pipelines
- Model comparison tooling for quality evaluation
- Centralized prompt management and versioning
E-Pharmacy Platform
E-commerce pharmacy with cart management, user accounts, order history, and ML-powered delivery time prediction.
OLX-Style Marketplace
B2C/C2C marketplace with dynamic category forms, bidding system with time extensions, NLP-powered search, and AI price prediction.
Textile Enquiry Portal
Industry-specific B2B enquiry management system for the textile sector, streamlining supplier-buyer communication and quote workflows.
ESOP & Rewards System
Employee stock ownership and rewards management platform with approval workflows, vesting schedules, and employee dashboards.
Real Estate Listings
Property listing platform with advanced search, filters, dynamic category-based forms, and location-aware browsing.
Cloud Infra & DevOps
End-to-end infrastructure design across AWS and Azure using Terraform. Includes WAF, security analysis, CI/CD pipelines, and GitHub Actions automation.
Writing
AI Blog
Technical deep-dives on LLMs, system design, and building AI-powered products in the real world.
Tracking the India Market During the Strait of Hormuz Crisis — April 2026
How crude at $115, a LICI bonus, ethanol policy, and defensive rotation shaped my trading week.
Building a Production LLM Gateway: Routing, Guardrails & Observability
How I built a self-hosted AI gateway that routes across GPT-4, Claude, and Gemini — with full cost tracking, content moderation, and trace-level observability using LiteLLM and Langfuse.
Natural Language to MongoDB Queries: How I Built It
A deep dive into building an NLP-to-query pipeline using LangChain and Gemini that lets non-technical users query a MongoDB database in plain English.
Ideas
AI Tools I'm Thinking About
Product ideas at the intersection of AI and real business problems. Some are concepts, some are being explored.
LedgerAI
Ask your accounts anything.
A natural language layer on top of accounting data. Business owners type questions like 'Who owes me the most this month?' and get instant answers — no dashboards needed.
DocGuard
Moderation infrastructure for AI-generated content.
A plug-and-play content moderation API for teams building LLM-powered apps. Combines rule-based guardrails with model-based classification to catch harmful outputs before they reach users.
SheetMind
Query your Excel files in plain English.
Upload a spreadsheet, ask a question, get an answer. Built for non-technical analysts who live in Excel but want the power of SQL-style queries without writing a single formula.
Contact
Let's build something.
Whether you're looking to hire, collaborate, or just want to talk about building AI-powered systems — my inbox is open.