TanvirulIslam.
Available for select engagements

Tanvirul Islam

Senior Software & AI Engineer

I build scalable software, AI-powered applications, autonomous AI agents, and production-ready machine learning solutions. 9+ years turning hard problems into reliable products.

Owner & BuilderExpense Copilot AI MCPAn AI-powered expense copilot that turns natural-language conversations into a simpler way to understand and manage spending.
  • Software Engineer
  • Generative AI Engineer
  • Machine Learning Engineer
  • AI Software Engineer
  • AI Software Developer
View ProjectsContact MeGitHubLinkedInDownload Resume

9+

Years experience

Find me on
  • ML
  • MCP
  • RAG
  • PYTHON
  • JAVA
  • LLM
  • BACKEND
  • FRONTEND
  • WEB
  • AGENTS
01About

Engineer across the full stack — web applications, backend systems, frontend interfaces, and autonomous AI.

Over 9+ years I've shipped production systems end to end — designing resilient backend architectures, then moving deep into applied AI: fine-tuned LLMs, retrieval pipelines, and autonomous agents that plan, reason, and call tools.

Today I focus on Model Context Protocol servers and AI-powered products — building the infrastructure that lets models act safely and reliably in the real world. I care about systems that hold up well beyond the demo.

Dhaka · Remote

  • Backend Architecture
  • AI Engineering
  • Machine Learning
  • MCP Servers
  • Autonomous Agents
  • Full Stack Software Engineer
02What I Do

Services.

AI Agent Development

Autonomous AI agents that plan, reason, and execute multi-step tasks — plus MCP servers that let models act safely and reliably in production.

  • LangGraph
  • MCP
  • Tool use
  • AWS Bedrock

LLM Fine-Tuning & RAG

Adapt models to your domain with LoRA, RLHF and DPO, and ground answers in your data with robust retrieval pipelines.

  • LoRA
  • RLHF/DPO
  • RAG
  • LangChain

Backend Development

Scalable, production-grade REST and GraphQL APIs, microservices and data layers built to handle real load — in Python, Java and Groovy.

  • FastAPI
  • Grails
  • Spring Boot
  • Django
  • REST/GraphQL

API Development

Design and build clean, well-documented REST and GraphQL APIs — with authentication, versioning, rate limiting and OpenAPI specs that other teams can rely on.

  • REST
  • GraphQL
  • OpenAPI
  • JWT/OAuth2
  • FastAPI

Frontend Development

Responsive, accessible UIs in Vue and React — from component libraries and design systems to full client-side applications.

  • Vue.js
  • React
  • JavaScript
  • TypeScript
  • CSS

Web Application Development

End-to-end web applications that tie together a clean frontend, robust backend, and database — delivered as production-ready products.

  • Full-Stack
  • Java
  • Python
  • MySQL
  • Docker
03Skills

A deep, current toolkit.

Languages

  • Java95%
  • Python95%
  • JavaScript90%
  • TypeScript85%
  • Groovy85%
  • SQL88%

Backend Development

  • Grails / Spring Boot92%
  • FastAPI / Django90%
  • Flask88%
  • REST / GraphQL90%
  • Microservices85%
  • WebSockets80%

AI Agent Development

  • MCP Server / Client92%
  • LangChain / LangGraph90%
  • AWS Bedrock Agents88%
  • Tool & Function Calling90%
  • Prompt Engineering88%
  • Multi-Agent Systems84%

AI & Machine Learning

  • LLM & RAG Pipelines92%
  • Fine-tuning (LoRA)88%
  • RLHF / DPO82%
  • PyTorch / TensorFlow82%
  • Scikit-Learn86%
  • Hugging Face88%

Frontend Development

  • Vue.js88%
  • React85%
  • HTML5 / CSS390%
  • TypeScript / ES6+85%
  • Responsive Design88%
  • Component Libraries82%

Cloud & DevOps

  • Docker88%
  • AWS85%
  • Kubernetes78%
  • GitHub Actions82%
  • Linux88%
04Featured Projects

Selected work.

  • Flagship AI Product

    Expense Copilot AI MCP

    Owned and built by Tanvirul Islam: an AI-powered expense copilot designed around conversational workflows and Model Context Protocol capabilities.

    • AI
    • MCP
    • Expense Intelligence
    • Product

  • ⬢ Agents

    Autonomous AI Agent Platform

    AI agents capable of planning, reasoning, and tool execution across multi-step tasks.

    • Python
    • LangChain
    • OpenAI
    • MCP

  • ▣ Production

    WebCommander CMS

    Enterprise website builder with native widgets, AI design matching, AI layout generation, and an AI deploy pipeline.

    • Java
    • Grails
    • JavaScript

  • ⌘ MCP

    MCP Server

    Production Model Context Protocol server that lets Claude and other AI hosts manage WebCommander stores directly — pages, components, publish/deploy and media across 20+ tools.

    • Python
    • FastMCP
    • OAuth2 PKCE
    • MCP

  • ◫ Production

    AI-Powered Website Builder & CMS

    Build it with AI, make it yours — an AI-driven website builder and CMS that generates pages and sections, then hands full editing control to the user.

    • Java
    • Grails
    • AI Agents
    • AWS Bedrock

  • ⊙ ML

    AI-Powered Fake Order Detector

    Real-time fraud detection model for a bookings platform, flagging fake orders as they arrive and improving continuously from new labeled data.

    • Python
    • Scikit-Learn
    • Real-time
    • Fraud Detection
05Process

How I work.

  1. 01

    Discover

    Understand the problem, users, constraints and what “good” actually means.

  2. 02

    Architect

    Design the systems, data flow and model approach — with evaluation built in from day one.

  3. 03

    Build

    Iterate fast with real feedback loops, tests and eval metrics guiding every step.

  4. 04

    Ship & Harden

    Deploy, monitor and refine until it holds up well beyond the demo.

06Experience

The journey so far.

  • Sep 2023 — Present

    Senior Software Engineer

    WebAlive · Remote

    • AI-powered CMS
    • MCP Server
    • AI Website Generation
    • Backend Architecture
    • AI Agents
    • Java
    • Python

  • Feb 2017 — Sep 2023

    Software Engineer

    Bit Mascot · Dhaka

    • Full-Stack
    • Grails
    • Java
    • Python
    • Editor Tooling
    • Performance
07Certifications

Credentialed in AI.

  • AI Engineer — Certified by micro1

    micro1 · Jul 2026 · AI Interview

  • Build AI Agents Using MCP

    IBM · Jun 2026 · MCP & Agents

  • Generative AI Advanced Fine-Tuning for LLMs

    IBM · Jun 2026 · LLM Fine-Tuning

  • IBM AI Developer — Professional Certificate

    IBM · Oct 2024 · Specialization

  • Building Generative AI-Powered Applications

    IBM · Oct 2024 · Generative AI

  • Developing AI Applications with Python & Flask

    IBM · Sep 2024 · AI Applications

  • Machine Learning with Python

    IBM · Nov 2024 · ML Foundations

  • Generative AI: Elevate Your Data Science Career

    IBM · Oct 2024 · Generative AI

  • Python for Data Science, AI & Development

    IBM · 2024 · Data Science

  • Generative AI: Prompt Engineering

    IBM · Sep 2024 · Prompt Engineering

08Open Source

Building in the open.

A portfolio of 16 open-source projects spanning LLM fine-tuning, RLHF, MCP & agents, generative-AI applications, and applied ML.

OSS projects
16

OSS projects

Focus areas
5

Focus areas

Years on GitHub
9+

Years on GitHub

  • Instruction Fine-Tuning with LoRA

    Parameter-efficient instruction tuning of a causal LM on CodeAlpaca-20k using LoRA via TRL's SFTTrainer, with completion-only loss masking and SacreBLEU evaluation.

    Python · LLM · RLHF

  • Reward Model Training (RLHF)

    GPT-2 + LoRA reward model trained on chosen vs. rejected preference pairs with TRL's RewardTrainer and a Bradley-Terry pairwise loss; evaluated by win rate.

    Python · LLM · RLHF

  • RLHF with PPO

    Fine-tunes GPT-2 with PPO using a sentiment classifier as the reward model and a KL-controlled reference policy — the reinforcement-learning stage of RLHF.

    Python · LLM · RLHF

  • Direct Preference Optimization (DPO)

    Aligns GPT-2 directly on (prompt, chosen, rejected) preference pairs using Hugging Face TRL's DPOTrainer + LoRA — no separate reward model needed.

    Python · LLM · RLHF

  • MCP Remote Suite

    Production-ready Model Context Protocol suite over Streamable HTTP, with a FastMCP server, filesystem roots security boundary, sampling, and an AI host that drives the tools.

    TypeScript · MCP · Agents

  • Permission-Aware MCP Security

    Server-enforced MCP permission management with per-tool risk classification, sandbox path confinement, single-use human approval, and tamper-resistant audit logging.

    Python · MCP · Agents

  • AI Writing Assistant Agent

    LangGraph agent that analyzes text and autonomously selects and chains writing-improvement tools, served through a Django web interface.

    Python · MCP · Agents

  • PDF Q&A Chatbot (RAG)

    Document-aware chatbot that answers questions over PDF content using LangChain and Flask, applying retrieval-augmented generation.

    Python · GenAI · NLP

  • AI Recommendation Engine

    eCommerce recommender combining collaborative and content-based filtering (Surprise, Scikit-learn) for real-time, personalized product suggestions.

    Python · Applied ML

09Tech Stack

Tools I reach for.

  • Python
  • Java
  • FastAPI
  • REST API
  • GraphQL
  • OpenAPI
  • Vue.js
  • React
  • Docker
  • TensorFlow
  • PyTorch
  • LangChain
  • LangGraph
  • OpenAI
  • AWS Bedrock
  • MCP
  • Kubernetes
  • Redis
  • MySQL
  • Grails
  • Spring Boot
  • JavaScript
  • TypeScript
  • HTML5/CSS3
  • Git
  • GitHub
10Writing

Latest articles.