Hazrat Ali

Applied AI & Machine Learning Engineer

LLM • RAG • AI Agents • Automation • Deep Learning • NLP • Research • Data Analytics

About Me

Applied AI and Machine Learning Engineer with hands-on experience in Machine Learning, Deep Learning, and Large Language Model (LLM)-based systems. Designing end-to-end ML and AI pipelines including data preprocessing, feature engineering, model training, evaluation, inference, deployment, and MLOps-driven workflow automation.

Experienced in building production-ready solutions such as Retrieval-Augmented Generation (RAG) systems, LLM-powered customer support automation, AI-driven document question answering, vector database–backed semantic search, and AI workflow automation using n8n for decision-support platforms based on transformer architectures.

Strong foundation in applied machine learning research with real-world business implementation.

Education

B.Sc. in Computer Science & Engineering
Green University of Bangladesh
CGPA: 3.40 / 4.00  |  2021 – 2025
Higher Secondary Certificate (HSC)
Hajigonj Model Government College
Chandpur, Bangladesh
Secondary School Certificate (SSC)
Payali K B M High School
Chandpur, Bangladesh

Experience

Click any card to expand details

AI / ML Engineer

9AM Solutions  •  2025 – 2026  •  On-site

Worked on building production-grade AI systems including LLM-powered agents, RAG pipelines, and intelligent automation workflows that reduced manual workload for business clients.

  • Built LLM-powered AI agents for customer support and order processing using RAG and prompt engineering
  • Developed and deployed predictive ML models for business forecasting and decision support
  • Designed end-to-end automation pipelines integrating ML models, LLM agents, and business logic
  • Built scalable FastAPI backend services for model inference and workflow orchestration
  • Reduced customer service workload by ~60% through intelligent automation
  • Built AI automation workflows using n8n, Zapier, and Make for end-to-end business process automation
  • Integrated automation tools including GoHighLevel, Crisp, Slack, and Airtable into client workflows

Research Assistant — ML & NLP

CRITS, Green University of Bangladesh  •  2024 – Present

Contributed to cutting-edge AI research on mental health detection, LLM-based systems, and NLP applications for low-resource languages including Bangla.

  • Conducted exploratory data analysis and preprocessing on large-scale social media datasets
  • Designed and implemented end-to-end deep learning and transformer-based experimental pipelines
  • Evaluated models using accuracy, precision, recall, and F1-score benchmarks
  • Co-authored multiple peer-reviewed journal and IEEE conference research papers
  • Worked on depression detection, suicide risk prediction, and RAG-based research generation systems

AI Developer

Softvence Agency  •  2026 – Present  •  On-site

Building AI-powered web platforms and intelligent automation solutions for business clients, integrating LLMs into real-world production environments.

  • Integrated LLMs into client-facing web platforms for smart interaction and content generation
  • Developed prompt engineering strategies for domain-specific AI behavior tuning
  • Built and deployed automated AI workflows for business process automation
  • Implemented RAG-based chatbots and customer support systems on production infrastructure
  • Managed deployment pipelines using Vercel, HuggingFace Spaces, and FastAPI
  • Designed AI automation solutions using n8n, Zapier, and Make for scalable business workflows
  • Integrated GoHighLevel, Crisp, Slack, and Airtable into automated AI-driven client systems

Skills

Programming

Python
SQL
C
C++

Machine Learning

Supervised Learning
Unsupervised Learning
Feature Engineering
Data Preprocessing
Model Training
Hyperparameter Optimization
Cross Validation
Model Selection
Evaluation Metrics
Accuracy
Precision
Recall
F1-score

Deep Learning

PyTorch
TensorFlow
ANN
CNN
RNN
LSTM
GRU
Transformers
Attention Mechanism

AI Agents & Automation

Autonomous AI Agents
Tool Calling
Task Planning
Multi-step Reasoning
Workflow Automation
Business Process Automation
Customer Support Automation
Order Processing Automation
n8n
Make AI
Zapier
Crisp
GoHighLevel
Slack
Airtable

Generative AI & NLP

Large Language Models (LLMs)
LangChain
Chains
Agents
Tools
Memory
HuggingFace Transformers
BERT
RoBERTa
DistilBERT
Prompt Engineering
Embeddings
RAG
Text Classification
NER
Summarization

Deployment & MLOps

FastAPI
REST APIs
Git
GitHub
Model Inference
Docker (Basic)
Vercel
HuggingFace Spaces

Data & Tools

Pandas
NumPy
Matplotlib
Scikit-learn
Git
GitHub
Linux
VS Code
Google Colab
Jupyter Notebook

Projects

Click any project to expand details

MediMate — AI Healthcare Assistant

Integrates symptom analysis, medical image understanding and LLM-powered conversation to provide early healthcare guidance.

ML • NLP • Computer Vision • LLM • FastAPI
MediMate addresses the challenge of accessible preliminary healthcare by combining three AI systems — a machine learning symptom checker, a CNN-based medical image analyser, and an LLM conversational assistant. Users get contextual, reasoned health guidance through a single interface. Deployed with FastAPI and Vercel for scalable real-time inference.
View on GitHub

Micro Tutor — AI Learning Assistant

Multilingual AI tutor supporting Bangla and English learners with adaptive, step-by-step explanations.

LLM • VLM • NLP • Python
Micro Tutor solves the problem of quality education access for Bangla-speaking students. It uses vision-language models to understand handwritten or image-based problems and generates adaptive explanations via LLMs. Students receive personalised, curriculum-aligned guidance in their native language.
View on GitHub

Bondhu — Mental Health Support AI

Emotion-aware conversational assistant providing bilingual mental health support using transformer-based NLP.

Transformers • NLP • LLM • FastAPI
Bondhu (meaning "friend" in Bangla) is built to bridge the mental health support gap in Bangladesh. The system detects user emotions in real time using transformer-based classification and generates empathetic, contextually appropriate responses. It supports both Bangla and English, making it accessible to a wider population.
View on GitHub

Medical Privacy RAG System

Privacy-preserving RAG chatbot that answers medical queries using only private, verified documents — zero data leakage.

LLM • RAG • Vector DB • LangChain
Built for healthcare organisations where patient data privacy is critical. The system grounds all LLM responses exclusively in ingested private medical documents using a vector database retrieval layer, ensuring the model never fabricates clinical information. Prevents hallucination and protects sensitive healthcare data.
View on GitHub

AI Customer Support Automation

End-to-end LLM pipeline that automates intent detection, knowledge retrieval, order handling and response generation.

LLM • RAG • LangChain • n8n • Automation
Designed for businesses with high customer query volumes. The pipeline automatically classifies intent, retrieves relevant knowledge via RAG, handles order lookups via API integration, and generates accurate replies — reducing human support workload by approximately 60% in production deployment.
View on GitHub

RAG Knowledge System — n8n Pipeline

Document-based AI knowledge system using PDF uploads, vector embeddings and Pinecone for intelligent retrieval.

RAG • Google Gemini • Pinecone • n8n
Users upload PDFs through a form. The system splits documents into smaller chunks, creates embeddings using Google Gemini, and stores them in Pinecone vector database. AI retrieves relevant document sections to answer questions accurately based on the uploaded knowledge base.
View on GitHub

LLM-Powered Customer Support Agent

Knowledge-base AI assistant using RAG, PDF embeddings and Pinecone vector search for intelligent customer support.

LLM • RAG • Google Gemini • Pinecone
Users upload PDF documents through a form trigger. The documents are loaded and split into smaller text chunks, converted into vector embeddings using Google Gemini, and stored in Pinecone vector database. AI later searches this knowledge base to answer customer questions accurately.
View on GitHub

GenAI Content Approval Pipeline

AI-driven multi-platform content generation and approval system for LinkedIn, Instagram, Facebook, X, TikTok and YouTube.

LLM • AI Agent • Content Generation • n8n
Collects a topic and requirements from a form, pulls related news from RSS feeds and Google News, scrapes articles and extracts key information. Uses an AI agent to score article relevance and generates platform-specific posts for LinkedIn, Instagram, Facebook, X, TikTok, and YouTube Shorts.
View on GitHub

AI-Powered Facebook Messenger Bot

LLM-based automated response system for Facebook page messaging using webhooks and AI agents.

LLM • AI Agent • Facebook API • Webhook
Uses a Webhook to receive events from Facebook Messenger, validates the request using hub.mode and verify_token, and extracts the user's message text from the incoming payload. Sends the message to an LLM-powered AI agent to generate an automated reply, enabling 24/7 customer engagement.
View on GitHub

Multilingual AI Restaurant Agent

LLM-powered restaurant assistant handling food orders and table bookings with automatic language detection.

LLM • AI Agent • NLP • Telegram
Triggered when a customer sends a chat message. The LLM detects the user's language and responds accordingly, handling food orders or table bookings. Collects user details (name, phone, items, time) and sends the confirmed order details to the restaurant owner via Telegram.
View on GitHub

LLM Inventory Management Agent

AI agent that interprets natural language to manage product inventory — check stock, update and track via chat.

LLM • AI Agent • Google Sheets • NLP
Starts when a chat message is received. An LLM-powered AI agent interprets the user request (check stock, update product, etc.), retrieves inventory data from Google Sheets, can append or update rows in the spreadsheet, and returns results back to the user via natural language chat.
View on GitHub

GenAI YouTube Video Creator

Automatically generate AI-powered video scripts, content and uploads to YouTube using Google Gemini.

GenAI • Google Gemini • YouTube API • LLM
Uses AI prompts to generate video ideas and scripts. Google Gemini generates short AI videos with auto-added titles and metadata. Uploads the generated video to YouTube using the YouTube API, designed mainly for short-form vertical video content.
View on GitHub

GenAI YouTube Shorts Idea Generator

AI-powered scheduled content ideation for YouTube Shorts using Google Gemini with structured JSON output.

GenAI • Google Gemini • AI Agent • Content
Runs on a scheduled trigger using an AI agent powered by Google Gemini. Generates short educational video ideas, structures output into JSON with title, hook, and learning content, ensuring each video idea is unique and beginner-friendly.
View on GitHub

Order Tracking Automation

Automated workflow that tracks orders in real time and sends status notifications via email and messaging platforms.

n8n • API Integration • Automation
Eliminates manual order tracking by integrating e-commerce APIs, databases and messaging services into a single automated workflow. Customers receive real-time updates on order status changes, reducing support queries and improving customer satisfaction.
View on GitHub

Customer Support Email Automation

AI-powered system that classifies, routes and auto-replies to customer emails using intent recognition.

LLM • n8n • Automation • Email
Processes incoming support emails automatically — classifying by topic, routing to the right department and generating contextual responses using LLM reasoning. Dramatically cuts response time and frees human agents for complex queries only.
View on GitHub

E-Commerce Inventory Management

Smart inventory monitoring system with automated low-stock alerts, order tracking and reporting dashboards.

n8n • APIs • Database • Automation
Automates the full inventory lifecycle for e-commerce businesses — monitoring stock levels, triggering reorder alerts, syncing order data across platforms, and generating daily inventory reports. Reduces stockout incidents and manual warehouse management effort.
View on GitHub

Census Geocoder + SerpAPI MAP

Converts addresses into geographic data and pulls census statistics using OpenStreetMap and U.S. Census APIs.

n8n • API Integration • Census API • OpenStreetMap
Takes an address input, uses OpenStreetMap API to convert it into latitude and longitude, then sends coordinates to the U.S. Census Geocoder API to find census tract, county, and state. Calls the ACS Census API to fetch demographic data and processes results in a code node to calculate totals and return structured output.
View on GitHub

Facebook Page Automation

Connects Facebook page messaging with an AI response workflow using webhooks and automated replies.

n8n • Webhook • Facebook API • AI Agent
Uses a Webhook to receive events from Facebook Messenger, validates the request using hub.mode and verify_token, and extracts the user's message text from the incoming payload. Sends the message to an AI agent to generate an automated reply, enabling 24/7 customer engagement on Facebook pages.
View on GitHub

Multi-step Content Approval System

Automates social media content creation and approval across LinkedIn, Instagram, Facebook, X, TikTok, and YouTube.

n8n • AI Agent • RSS • Content Automation
Collects a topic and requirements from a form, pulls related news from RSS feeds and Google News, scrapes articles and extracts key information. Uses an AI agent to score article relevance and generates platform-specific posts for LinkedIn, Instagram, Facebook, X, TikTok, and YouTube Shorts.
View on GitHub

AI Customer Support Email Agent

Knowledge-base AI support assistant using PDF documents, vector embeddings and Pinecone for intelligent responses.

n8n • Google Gemini • Pinecone • RAG
Users upload PDF documents through a form trigger. The documents are loaded and split into smaller text chunks, converted into vector embeddings using Google Gemini, and stored in Pinecone vector database. AI later searches this knowledge base to answer customer questions accurately.
View on GitHub

Inventory Management AI Agent

Manage product inventory using AI chat — check stock, update products and track inventory via Google Sheets.

n8n • AI Agent • Google Sheets • Automation
Starts when a chat message is received. An AI agent interprets the user request (check stock, update product, etc.), retrieves inventory data from Google Sheets, can append or update rows in the spreadsheet, and returns results back to the user via chat for seamless inventory management.
View on GitHub

Restaurant Management Assistant

AI assistant for food orders and table bookings with multilingual support and Telegram notifications.

n8n • AI Agent • Telegram • Multilingual
Triggered when a customer sends a chat message. AI detects the user's language and responds accordingly, handling food orders or table bookings. Collects user details (name, phone, items, time) and sends the confirmed order details to the restaurant owner via Telegram.
View on GitHub

RAG in n8n — Document Knowledge System

Build a document-based AI knowledge system using PDF uploads, embeddings and Pinecone vector search.

n8n • RAG • Google Gemini • Pinecone
Users upload PDFs through a form. The system splits documents into smaller chunks, creates embeddings using Google Gemini, and stores them in Pinecone vector database. AI retrieves relevant document sections to answer questions accurately based on the uploaded knowledge base.
View on GitHub

YouTube Video Automation

Automatically generate and upload AI-powered videos to YouTube with titles, scripts and metadata.

n8n • Google Gemini • YouTube API • Automation
Uses AI prompts to generate video ideas and scripts. Google Gemini generates short AI videos with auto-added titles and metadata. Uploads the generated video to YouTube using the YouTube API, designed mainly for short-form vertical video content.
View on GitHub

Custom Puppeteer Login Automation

Automatically log into websites using email links with Gmail monitoring, Apify browser automation and data scraping.

n8n • Apify • Gmail • Puppeteer
Watches Gmail inbox for specific messages, extracts login links from email text, and sends the login URL to Apify automation. Apify runs a browser automation script to log into the platform and returns scraped data from the authenticated session.
View on GitHub

YouTube Shorts Idea Generator

Automatically generate unique, beginner-friendly YouTube Shorts content ideas on a schedule using AI.

n8n • Google Gemini • AI Agent • Scheduling
Runs on a scheduled trigger using an AI agent powered by Google Gemini. Generates short educational video ideas about Adobe tools, structures output into JSON with title, hook, and learning content, ensuring each video idea is unique and beginner-friendly.
View on GitHub

Typeform → Duplicate-Safe Credits System

Manage user submissions and credit balance with duplicate detection, payment validation and Google Sheets sync.

n8n • Typeform • Google Sheets • Automation
Triggered when a Typeform form is submitted. Compares the user with existing records in Google Sheets, detects duplicate users by matching name and email, updates or appends the user's credit data, validates payments and updates the user's remaining credits automatically.
View on GitHub

Shopify Product Metadata Extractor

Extract product data and metafields from Shopify using GraphQL API with pagination and scheduled triggers.

n8n • Shopify API • GraphQL • Automation
Runs on a scheduled trigger, sends a GraphQL request to Shopify API, retrieves products and their metafields, and splits products into individual items for processing. Handles pagination until all products are collected for complete data extraction.
View on GitHub

Shopify Data Sync Workflow

Sync Shopify products and variants for inventory analysis with automated data cleaning and scheduling.

n8n • Shopify • Data Processing • Automation
Runs on a scheduled daily/weekly trigger. Fetches products from a Shopify store, splits each product into variants, cleans descriptions and tags, and processes data for inventory or content analysis workflows.
View on GitHub

Fake News Detection System

ML and transformer-based classifier that identifies misinformation articles with high accuracy.

ML • NLP • BERT • Transformers
Tackles the growing problem of online misinformation by combining traditional ML classifiers with fine-tuned BERT models. Trained on multi-source labelled news datasets, the system achieves high precision in distinguishing real from fabricated news content and is applicable to social media monitoring.
View on GitHub

Blood Donation Management System

NLP-powered platform where users find blood donors using natural language queries with location extraction.

NLP • Web • Database
Addresses emergency blood availability challenges by allowing users to search for donors using plain language — e.g. "O+ donor near Dhaka tonight". NLP extracts blood type, urgency and location from queries to return the most relevant donor matches from the database.
View on GitHub

Depression Detection System

Deep learning model predicting depression risk from Bangla social media posts with transformer-based analysis.

DL • NLP • Transformers • Bangla
Developed as part of published IEEE STI 2025 research. Analyses Bangla social media text to identify linguistic markers of depression and suicidal risk at an early stage. The model uses transformer-based architectures fine-tuned on a custom labelled Bangla dataset, enabling early intervention support.
View on GitHub

Crop Disease Detection System

CNN-based image classifier that identifies crop diseases from leaf photos to help farmers take early action.

CNN • TensorFlow • Image Processing
Designed to support Bangladeshi farmers with limited access to agricultural experts. Users photograph a crop leaf and the CNN model identifies the disease type and severity. Early detection enables timely treatment, reducing crop loss and improving agricultural yield.
View on GitHub

AI Object Detection System

Real-time object detection system processing images and live video streams using YOLO and CNN architectures.

Computer Vision • YOLO • CNN • OpenCV
Implements state-of-the-art YOLO models for high-speed, accurate object detection in both static images and real-time video. Applicable to surveillance, inventory counting, and industrial quality control. Built with OpenCV for efficient video stream processing.
View on GitHub

Suicide Risk Prediction

Transformer-based early warning system identifying suicidal risk signals in Bangla text data.

Transformer • NLP • Deep Learning
An extension of the depression detection research, this system focuses specifically on identifying high-risk suicidal ideation markers in online posts. Fine-tuned transformer models are used to classify severity levels, enabling mental health professionals and platforms to prioritise at-risk individuals for intervention.
View on GitHub

University Management System

Full-featured university ERP platform with student, teacher and admin dashboards, attendance, results and registration.

HTML • CSS • JavaScript • PHP • MySQL
A complete academic management platform built to streamline university operations. Features include student enrollment and course registration, attendance tracking, grade management, teacher assignment, and an admin dashboard for institutional oversight — all in one integrated system.
View on GitHub

Website Customer Support Chatbot

Automated chatbot embedded on websites to handle visitor queries, FAQs and support requests around the clock.

n8n • Chatbot • Automation • LLM
A deployable chatbot solution for business websites that handles frequently asked questions, product enquiries and support tickets without human intervention. Built with n8n workflows and LLM response generation, it integrates seamlessly into any existing website and operates 24/7.
View on GitHub

Publications

Click any paper to expand details

Deep Learning-Based Early Prediction of Depression & Suicidal Risk in Bangla Social Media Posts — IEEE STI 2025
Presents a transformer-based deep learning framework trained on a custom Bangla social media dataset to detect early-stage depression and suicidal ideation. Combines BERT-based fine-tuning with multi-class risk classification to assist mental health intervention systems. Accepted and presented at IEEE STI 2025.
✓ Accepted — IEEE STI 2025
Transformer-Based Hybrid Framework for Early Detection of Depression and Suicidal Risk — Journal
Proposes a hybrid architecture combining multiple transformer models with ensemble decision logic for improved accuracy in detecting mental health risk from text. Evaluates the system across multiple Bangla and English datasets, demonstrating significant performance improvements over baseline models.
⏳ Under Review — Journal
Retrieval-Grounded and Novelty-Aware Framework for Research Idea Generation with Large Language Models — Journal
Introduces a RAG-based system that assists researchers in generating novel research ideas by grounding LLM outputs in existing literature and enforcing a novelty constraint. The framework reduces redundancy and hallucination in AI-assisted research ideation, evaluated on academic paper repositories.
⏳ Under Review — Journal
Enhancing Reinforcement Learning Alignment via LLM-Based Interpretation of Human Natural Language Feedback — IEEE QPAIN 2026
Explores using Large Language Models as an intermediary to interpret and structure ambiguous human feedback for reinforcement learning from human feedback (RLHF) pipelines. The approach improves reward signal quality and alignment stability in fine-tuning scenarios.
✓ Accepted — IEEE QPAIN 2026
A Large Language Model–Driven Reinforcement Learning Framework for Intelligent Employee Performance Optimization — IEEE QPAIN 2026
Proposes an LLM-driven reinforcement learning framework that models and optimizes employee performance through intelligent feedback interpretation, adaptive goal setting, and personalized development strategies in organizational environments.
✓ Accepted — IEEE QPAIN 2026

Achievements & Certificates

Click any certificate to expand details

Experience Certificate — AI/ML Developer 9AM Solutions

Officially certifies employment as an AI/ML Developer at 9AM Solutions, recognising hands-on work in building LLM-powered agents, RAG systems, and automation workflows for real business clients.

IEEE STI 2025 — Certificate of Appreciation IEEE

Awarded in recognition of authoring and presenting an accepted research paper on deep learning-based mental health detection at the IEEE STI 2025 international conference.

Master in Python Language Certification CampusX

Comprehensive Python programming certification covering core syntax, data structures, OOP, file handling, and Python for data science applications.

Master in Machine Learning Certification CampusX

Hands-on training covering supervised and unsupervised learning algorithms, feature engineering, model evaluation metrics, hyperparameter tuning and real-world ML pipeline construction.

Master in Deep Learning Specialization CampusX

In-depth specialisation in neural network architectures including ANN, CNN, RNN, LSTM and GRU, with practical implementation using TensorFlow and PyTorch on real-world datasets.

Master in NLP Certification CampusX

Full NLP curriculum covering text preprocessing, tokenisation, embeddings, transformer architectures, BERT fine-tuning, sentiment analysis and production language model deployment.

LLM & RAG Systems Training CampusX

Advanced training on Large Language Models, vector databases, embedding strategies, LangChain, and building production-grade Retrieval-Augmented Generation pipelines.

AI Agent & Automation Certification Vibe Academy

Practical certification in building autonomous AI agents using tool calling, task planning, multi-step reasoning and workflow automation frameworks including n8n and Make AI.

AI Agent & Automation Certification Ostad

Certification in designing and deploying autonomous AI agents, integrating APIs, building automated business workflows, and implementing end-to-end AI-driven business process automation.

Computer Vision Bootcamp Bootcamp Program

Intensive bootcamp covering image classification, object detection with YOLO, semantic segmentation, and OpenCV-based real-time video processing applications.

Wadhani Foundation — Professional Skills Wadhani Foundation

Professional development certification covering workplace communication, leadership, teamwork, problem-solving and professional ethics — skills for effective collaboration in tech environments.

Contact

Feel free to reach out for collaboration, research opportunities, or project enquiries. I'm always open to interesting AI and ML work.

📍 Mirpur 11, Dhaka, Bangladesh
📧 hazratali26725@gmail.com
📞 +880-1883-860-803
Download My CV