COMPOSING IDENTITY
APPLICATION DEVELOPER(AI/ML)DHAKA, BANGLADESH

Mehedi Hasan Nipu.

AI engineer & researcher

I build reliable agentic systems, language-model applications, and applied machine-learning products for real operational work.

Portrait of Mehedi Hasan Nipu
MEHEDI HASAN NIPU
Scroll to profile

01 / PROFILE

From research questions to software that works in the real world.

I am an AI engineer and researcher focused on building useful, accountable systems around large language models.

At DEXIAN, I design agentic workflows with the Claude Agent SDK, LangGraph, and Azure OpenAI. My work spans government procurement, software generation, enterprise compliance, and model-cost analytics. Across each domain, the goal is the same: make advanced AI dependable enough to use.

My research extends that practice into trustworthy AI, multi-agent collaboration, medical decision support, and human-LLM interaction.

06Production systems selected below
12Research outputs in the dedicated archive
03Years bridging research and delivery

02 / PROFESSIONAL EXPERIENCE

Roles, projects, and the work behind them.

Each project is grouped under the position where I designed, built, and shipped it.

Application Developer (AI/ML)

Building agentic AI products, multi-agent development systems, and cost-aware model infrastructure for production business workflows.

01

AuditPilot: Multi-Agent AI Platform for Regulatory Compliance Auditing

  • Built a Control Set builder that converts any regulatory standard, including GAMP 5, ISO, FDA, and internal SOPs, into a machine-readable clause index using quality-scored PDF/DOCX extraction and parallel Claude Agent SDK workers.

  • Engineered a clause-by-clause compliance checker where a scoping agent bounds document responsibility and concurrency-capped agents return evidence-cited verdicts with Critical, Major, and Minor severity grading.

  • Designed the pipeline for resilience with per-slice checkpointing to Azure Blob Storage for resumable retries, citation verification against source files, and deterministic Python-based compliance scoring.

  • Delivered a grounded chat interface for instant, cited Q&A over each Control Set, plus a unified activity and cost dashboard tracking agent spend and status in Azure SQL.

TECH STACK
  • Python 3.12
  • FastAPI
  • Next.js
  • React
  • TypeScript
  • Claude Agent SDK
  • Anthropic Claude Opus/Sonnet/Haiku
  • pdfplumber
  • python-docx
  • Azure SQL
  • Azure Blob Storage
  • Docker
  • Azure App Service
02

RFPGen: AI-Powered Government RFP Aggregation & Proposal Generation Platform

  • Developed a SAM.gov scraper and multi-portal aggregation pipeline to automatically discover ServiceNow and ITSM opportunities across US federal and SLED procurement portals, with keyword-based relevance scoring and structured Excel output.

  • Integrated Azure Document Intelligence and GPT-4.1 to analyze RFP documents, extract structured requirements, and enable natural-language Q&A over indexed document sections stored in Azure SQL.

  • Developed Claude Agent SDK-powered agents for automated company profiling and end-to-end DOCX and PPTX proposal generation tailored to specific RFP requirements.

  • Built a unified activity logging and cost analysis dashboard tracking user interactions, agent invocations, document operations, and real-time API spend across Anthropic Claude and Azure OpenAI, with full audit trails in Azure SQL.

TECH STACK
  • Python 3.12
  • FastAPI
  • React 18
  • TypeScript
  • LangGraph
  • LangChain
  • Claude Opus/Sonnet
  • Azure OpenAI GPT-4.1
  • Azure Document Intelligence
  • Azure SQL
  • Azure Blob Storage
  • Selenium
  • Patchright
  • Firecrawl
03

BM Product Accelerator: AI-Powered Full-Stack Application Generator

  • Developed a multi-agent orchestration system using Claude Agent SDK to analyze PRD documents and generate implementation plans by discovering reusable components in Azure Blob Storage.

  • Engineered an automated code-generation pipeline with BackendBuilderAgent creating validated FastAPI applications and FrontendBuilderAgent scaffolding React and Vite projects with npm management.

  • Implemented a skill-generation system that extracts reusable patterns from GitHub repositories or code archives, generates documentation, and updates the component library.

  • Built an interactive natural-language CLI for conversational project modification and debugging with automatic port management and session persistence.

  • Architected an asynchronous FastAPI backend with Azure Blob Storage, multi-format document parsing for PDF, DOCX, and Markdown, and Pydantic validation.

TECH STACK
  • Claude Agent SDK
  • FastAPI
  • React
  • Vite
  • Azure Blob Storage
  • Pydantic
  • Multi-Agent Systems
  • Natural Language Processing

Junior Application Developer (AI/ML)

Shipped Salesforce automation, document-intelligence workflows, computer-vision fraud detection, and geospatial compliance products.

04

SFAGent: AI-Powered Salesforce Agentic Compliance System

  • Created Salesforce Apex classes to handle POST and GET operations and enable natural-language interaction with Salesforce data.

  • Developed custom authentication functions to securely connect Apex classes and manage efficient API communication.

  • Integrated an Azure AI Foundry Agent to automate Salesforce workflows from natural-language instructions.

  • Developed text and speech-to-text FastAPI services using Azure OpenAI Whisper for accurate audio transcription.

  • Built an intuitive multimodal frontend for text and audio operations with an automatic agentic flow.

TECH STACK
  • Salesforce Sandbox
  • Apex Class
  • Azure AI Foundry
  • GPT-4.1
  • Whisper
  • FastAPI
05

AI-Powered Automated Check Fraud Detection

  • Developed an AI-based check fraud detection system using advanced machine-learning techniques to identify fraudulent checks.

  • Leveraged Azure Document Intelligence and the prebuilt US Bank Check model to validate critical check components.

  • Trained a Siamese neural network for matching signatures against reference and blacklist samples.

  • Designed a three-level review classification categorizing check images as Safe, Suspicious, or Fraud.

TECH STACK
  • Python
  • Azure Document Intelligence
  • Siamese Neural Network
  • React
  • Django
  • TensorFlow
06

GeoViz: Modern Flight-Based Compliance System

  • Developed a responsive geographic analysis application using Next.js, React, and shadcn/ui for coordinate and compliance workflows.

  • Integrated real-time geographic tracking with Leaflet.js maps for dynamic coordinate visualization.

  • Implemented coverage-analysis tools using the Haversine formula for accurate distance calculation.

  • Designed a responsive UI system optimized for desktop, tablet, and mobile devices.

TECH STACK
  • Python
  • Next.js
  • React
  • Leaflet.js
  • shadcn/ui
  • Tailwind CSS
  • Lucide React
  • GeoJSON
  • Haversine Formula
  • Axios
  • Django

AI & Machine Learning Researcher

Developed RAG agents, competitive-intelligence workflows, and document-processing applications while evaluating retrieval quality.

07

GenBot: Advanced RAG-Based Web Chatbot with CrewAI

  • Developed a vectorization and storage pipeline for unstructured JSON data using OpenAI embeddings and ChromaDB for fast retrieval.

  • Implemented RAG-based agents combining vector retrieval and external web search for dynamic chatbot responses.

  • Designed multi-source knowledge refinement to improve response relevance and accuracy in real time.

  • Evaluated embedding models with RAGAS to select the most effective model for the target workload.

TECH STACK
  • Python
  • OpenAI
  • ChromaDB
  • LangChain
  • React
  • Google Custom Search API
  • WebCrawler
  • Playwright
08

Company Competitor Service Analysis and SWOT AI Agent

  • Developed an AI-powered competitive-analysis system for extracting company services and identifying competitors through web scraping and OpenAI models.

  • Built a comprehensive SWOT workflow using dynamic AI responses grounded in company and competitor web content.

  • Integrated extraction, analysis, and report generation to deliver structured insights into market positioning and the competitive landscape.

TECH STACK
  • Python
  • OpenAI
  • Streamlit
  • BeautifulSoup
  • Pandas
  • GPT-4
  • CrewAI
09

LLM Powered Legal Contract Generator

  • Developed a Flask-based contract-generation application using AI and OCR to extract, refine, and generate legal document text.

  • Integrated Groq and LLaMA Vision for accurate processing of complex legal text and images.

  • Incorporated OCR document processing for digitizing scanned and image-based legal contracts.

TECH STACK
  • Python
  • Flask
  • Groq API
  • LLaMA Vision
  • PyMuPDF
  • OpenCV

03 / EDUCATION

The foundation beneath the practice.

NORTH SOUTH UNIVERSITY

B.Sc. in Computer Science & Engineering

January 2019 - January 2024 · Dhaka, Bangladesh

Thesis
Comparative Analysis of Deep Learning Algorithms for Multiple Disease Prediction
Focus
Machine learning, deep learning, computer vision, and healthcare AI
Credits
138

BIRSHRESHTHA NOOR MOHAMMAD PUBLIC COLLEGE · BNMPC

Higher Secondary Certificate (HSC), Science

Jun 2016 - Jun 2018 · Dhaka, Bangladesh

Mehedi Hasan Nipu at his North South University graduation
GRADUATION / 2024North South UniversityDhaka, Bangladesh

04 / TECHNICAL SKILLS

The working stack behind the systems.

Languages, frameworks, cloud services, and engineering practices used across production applications and research.

01

Programming Languages

  • Python (NumPy, scikit-learn, Matplotlib, Pandas, Seaborn)
02

Web Development

  • HTML5
  • CSS3
  • FastAPI
  • Flask
  • React JS
  • Next.js
  • Streamlit
03

Database

  • MySQL
  • Azure SQL
04

Deep Learning Frameworks

  • TensorFlow
  • Keras
  • PyTorch
05

LLM Application Frameworks

  • LangChain
  • LangGraph
  • LangSmith
  • Ragas
  • DeepEval
  • CrewAI
  • OpenAI
  • Claude Agent SDK
06

Cloud Services

  • Azure OpenAI
  • Azure AI Foundry
  • Azure App Service
  • Azure Blob Storage
  • Docker
  • Azure Boards
07

Others

  • Vector Database
  • OpenCV
  • GitLab
  • GitHub
  • GitHub Copilot
  • Linux
  • LLM Fine-tuning
  • Architecture & Data Flow Diagrams
  • Data Scraping
  • Data Annotation
  • Data Analysis
  • Feature Engineering

05 / AWARDS & CERTIFICATIONS

Awards & Certifications.

Verified credentials, professional service, and recognition from academic and industry organizations.

06 / FEATURED PUBLICATIONS

Selected research.

Three recent outputs from the complete research archive.

Preprint16 Aug 2026

Agentic Security: A Systematization of Tools, Failure Modes, and Design Laws for LLM-Driven Penetration Testing

Israt Moyeen Noumi, Tarannum Ahmed Nowshin, Md. Mehedi Hasan Nipu, Mohammad Sakib Mahmood, Md. Jakir Hossain, M. F. Mridha

Preprint4 Aug 2026

Robust and Personalized Federated Learning for Aircraft-Engine Prognostics under Benign and Adversarial Client Heterogeneity

Chinmoy Mitra, Md Mehedi Hasan Nipu, Mohammad Sakib Mahmood, Md Rakibul Islam, MF Mridha

Preprint20 Jul 2026

ChannelGuard: Safe Models Do Not Compose into Safe Multi-Agent Systems

Elias Hossain, Md Mehedi Hasan Nipu, Fatema Tuj Johora Faria, Tasfia Nuzhat Ornee, Maleeha Sheikh

07 / NOW & RECENT

Field notes from the work.

July 2026

Field note

Published Top-K Starvation: A Retrieval-First Fix, a practical account of restoring RAG quality by fixing recall before changing the model.

May 2026

Field note

Published IPC Overflow: A Data-First Fix, documenting a production lesson from high-volume agentic application pipelines.

08 / CAPABILITIES

Where I can contribute.

A

Agentic AI engineering

Production-grade agent orchestration, RAG systems, evaluation, tracing, and model-cost observability.

B

Applied machine learning

Computer vision, clinical AI, document intelligence, fraud detection, and trustworthy model design.

C

Full-stack AI products

FastAPI and React applications that turn complex model capabilities into clear, reliable user workflows.

HAVE A HARD AI PROBLEM?

Let's make it useful.

I am open to research collaboration, applied AI projects, and conversations about agentic systems that need to work beyond the demo.

Start a conversation