Generative AI Engineer | AI Agents with AutoGen & LangChain | RAG Architect | MCP Developer | Exploring NLP, Neural Netw

leetcode

github

Hemant Kumar  • Peerlist

• Designed a RAG pipeline with multiple chunking strategies (fixed-size, recursive, semantic, and custom smart chunking) for context-rich document retrieval. • Smart chunking improved retrieval relevance and reduced context overlap, boosting answer precision. • Integrated the pipeline with large language models for efficient, domain-specific querying

Retrieval-Augmented Generation (RAG) Applicat

•Designed and developed a full-stack platform with both a web-based interface and REST API for automated extraction, OCR, and KPI processing of diverse bank statement PDFs (including scanned statements). • Implemented modular backend pipelines using Tabula (Java), Mistral OCR, and YAML configurations to support multiple banks with flexible, reusable processing logic. • Integrated AI-driven and rule-based bank auto-detection in the frontend, enabling seamless backend pipeline selection without manual input. • Built advanced CSV/Excel utilities for cleaning, merging, column alignment, and format conversion, reducing manual processing effort by 90%. • Attended client meetings to gather requirements and translate them into actionable development features under tight delivery deadlines.

Unified Bank Statement Processing Platform

• Developed a modular automation tool to interact with Google Drive, supporting listing, reading, and traversing files/folders with advanced recursive operations and MIME type filtering. • Engineered a robust OAuth2 authentication flow with token refresh and persistent credential storage for seamless, uninterrupted API access. • Built a command-line server controller (mcp server.py) to orchestrate Drive operations, cache metadata locally, and provide modular utility scripts for specific actions (e.g., file search, folder tree visualization). • Designed a scalable, maintainable architecture with separated concerns, reusable components, and structured logging for improved debugging and reliability.

Google Drive MCP Automation Tool

• Built a multi-agent architecture for automated coding workflows: Python, Automation, AI Agents– Code Agent– Scrapes coding problems (e.g., LeetCode) and generates Python solutions.– Test Case Generator Agent– Creates diverse test cases.– Execution & Validator Agent– Runs solutions and checks them.– Critic Agent– Analyzes failures and suggests fixes. • Reduced debugging cycles by automating error detection and correction loops.

Multi-Agent Coding System

• Developed a full-stack event management platform with secure JWT-based authentication, role-based access, and real-time event updates. • Features include event creation, editing, deletion, categorization, and filtering for efficient management. • Improved user retention through a modern, responsive UI and intuitive navigation

Meeting & Event Scheduling Platform

• Developed a full-stack food delivery application using MERN stack technologies (MongoDB, Express JS, React JS, Node JS)
 to provide users with a seamless browsing and ordering experience.
 • Implemented user and admin functionalities, including order placement, cart management, and an admin interface for
 menu and order management, enhancing operational efficiency. .
 • Designed a modern, user-friendly interface with HTML5, CSS3, and Bootstrap, featuring a clean, minimalist design for
 intuitive navigation and ease of use.

Food Delivery App

• Developed an automated subjective answer evaluation system using machine learning and NLP techniques.
 • Implemented cosine similarity to compare student answers with model answers, achieving 87% accuracy
 • Reduced evaluation time and enhanced grading consistency, providing immediate feedback for educational assessments

Beyond Rubrics

• Created the Toolease app using XML and Java, streamlining the online buying and selling of tools
 • This innovative platform has led to a remarkable reduction in human effort, significantly boosting operational efficiency in
 the tool market
 • Users can now easily connect, trade, and save valuable time in their tool transactions

ToolEase

* Built a React-based Notes App that allows users to create and manage note groups with unique names for seamless organization and data persistence.
* Users can add, edit, and save notes within groups, with data stored in browser storage to ensure notes persist across page reloads.
* Features an intuitive UI with dynamic group icons, automatic popup handling, and metadata tracking for note creation and updates.

React notes

Built an interactive Rock Paper Scissors game with a clean and modern UI. The game allows users to challenge a computer opponent, keeping track of scores in real-time. It features intuitive gameplay, smooth animations, and a responsive design for an engaging experience.