🤖 Ha ha ha… who knew one "Hello World!"
would lead to building intelligent systems
powered by AI
and LLMs?
I'm Sonu Kumar, an AI enthusiast turned engineer who evolved from debugging C
programs to
developing advanced AI-driven applications. My journey began with curiosity and has grown into a
passion for
solving real-world problems using cutting-edge technologies.
I specialize in creating smart systems with Large Language Models,
LangChain, LangGraph, and RAG architectures.
One of my key
projects, DB-Agent, translates natural language into executable SQL using LLMs —
making
database access conversational and intuitive.
In another project, I used OCR (EasyOCR) and YOLOv8 to detect and
mask
Aadhar numbers from images, addressing data privacy through AI and computer vision. These solutions
showcase
my ability to combine NLP, vision, and automation to create secure, scalable tools.
On the backend, I build robust APIs using Java Spring Boot to support these AI
pipelines and
integrate them with external platforms seamlessly. I focus on clean architecture, scalable design,
and
performance optimization.
When I'm not debugging, I’m experimenting with model fine-tuning and agent workflows — because
nothing excites
me more than turning complex ideas into working AI prototypes.
Let’s build AI systems that are smart, helpful, and maybe a little funny too. 😄
As part of Monocept, I engineered a sophisticated LLM-based agent designed to bridge the gap between natural language and database queries. This agent translates complex user questions (e.g., "Show me all customers who ordered more than 5 items last month") into precise MySQL queries, eliminating the need for users to have SQL knowledge.
Key Technologies: Leveraged LangChain for robust agent orchestration and prompt engineering, integrating with OpenAI's powerful language models for natural language understanding and SQL generation. The backend was built with Python, interacting directly with MySQL databases.
Impact: This project significantly improved data accessibility and operational efficiency for non-technical users across various departments. It reduced dependency on database administrators for routine queries, enabling faster insights and decision-making, and automating data retrieval workflows that previously required manual SQL scripting.
I developed a secure and automated pipeline for redacting sensitive Aadhaar numbers from various uploaded documents. This critical project addressed data privacy concerns by ensuring that personal identification numbers were automatically detected and masked, preventing unauthorized exposure.
Key Technologies: The pipeline utilized Optical Character Recognition (OCR) for accurate text extraction from documents. I fine-tuned YOLOv8, a state-of-the-art object detection model, to achieve high accuracy in identifying the precise location of Aadhaar numbers within diverse document layouts. The entire system was implemented in Python, leveraging OpenCV for image processing tasks.
Impact: This solution drastically reduced manual effort and potential human error in handling sensitive data. It significantly enhanced data security and compliance with privacy regulations, providing a scalable and efficient method for document processing within the organization.
mPro is a web application designed for insurance agents to efficiently fill in user details and facilitate the purchase of insurance policies. I played a crucial role in the backend development and maintenance of this existing project. My responsibilities included identifying and resolving critical bugs, ensuring the application's stability and reliability. I also developed and integrated new Spring Boot APIs to extend the application's functionality and support evolving business requirements.
Key Technologies: The project heavily relied on Java and the Spring Boot framework for building robust and scalable microservices. A significant aspect involved seamless integration with AWS S3 for secure, highly available, and cost-effective storage of various application assets and user data.
Impact: My contributions directly led to improved system stability and enhanced user experience by resolving performance bottlenecks and adding new features. The secure integration with AWS S3 ensured data integrity and scalability, supporting a growing user base and critical business operations for MaxLife Insurance.
I spearheaded a critical database migration initiative for Care Health Insurance, overseeing the complex transition from their legacy Oracle database system to a more modern and scalable PostgreSQL environment. This project involved comprehensive planning, data mapping, and rigorous testing to ensure data consistency and integrity throughout the migration.
Key Technologies: The migration involved extensive use of Java for developing custom scripts and tools to facilitate data extraction, transformation, and loading (ETL) processes. Crucially, this included refactoring and adapting existing Java Spring-based project queries to be compatible with the new PostgreSQL database. Deep understanding of both Oracle and PostgreSQL database architectures was essential to ensure a seamless transition.
Impact: My leadership ensured the successful completion of the migration with zero-downtime, minimizing disruption to critical business operations. This modernization effort resulted in significant improvements in database performance, scalability, and cost-efficiency, laying a robust foundation for future data-driven initiatives at Care Health Insurance.
MoviesBot is a responsive chatbot application designed to enhance user engagement by providing relevant answers to movie-related questions. It leverages a Retrieval Augmented Generation (RAG) pipeline, utilizing Pinecone as its vector database to efficiently retrieve information from a pre-defined knowledge base and generate accurate, contextually rich responses.
This is an AI-powered Spring-related chatbot where users can inquire about topics such as Spring, Spring Boot, and Java. Additionally, users have the option to provide feedback for the service.
A simple Electricity application written in Java which allows customers to create an account, bill and payment, and bill status.
EasyRentHub is a housing system application that allows users to view and manage property rentals.
The Online Trip Management System is a web application that aims to provide a convenient platform for managing tours and travels.
Purplle is a prominent online beauty and personal care platform, through the implementation of cutting-edge technologies and strategic improvements.
Made with ❤ By Sonu Kumar