Computer Science graduate from Brac University, Dhaka building production-grade AI and full-stack systems.

I've shipped an AI-powered hospital management system combining Gemini LLM and Random Forest classifiers, an NLP pipeline for sentiment analysis and fake review detection on Amazon datasets, and a full-stack food ordering platform. Currently exploring neural style transfer (PyTorch / VGG19) and distributed parallel computing.

I'm looking for AI engineering, backend, or ML roles — internships or full-time — where I can contribute real impact from day one.

Python · PyTorch · VGG19 · AdaIN · Flask · Computer Vision

Built an end-to-end neural style transfer web application from scratch using Adaptive Instance Normalization (AdaIN) — a real-time feed-forward approach that decouples style from content without per-image optimization.

• Trained a custom decoder network paired with a frozen VGG19 encoder, optimizing a weighted combination of perceptual content loss and Gram matrix-based style loss across multiple feature layers
• Implemented AdaIN normalization to align channel-wise mean and variance of content features with style features, enabling arbitrary style transfer at inference time
• Deployed as a Flask web app with adjustable style strength (alpha blending) and real-time image preview

React · Flask · Gemini LLM · Scikit-learn · PostgreSQL · JWT · RBAC

A modular, role-based healthcare platform for Admin, Doctor, Patient, and Receptionist workflows built on a decoupled REST architecture.

• Dual-AI pipeline: Random Forest classifiers for clinical risk prediction + Gemini LLM for diagnostic reasoning and automated SOAP note generation
• Drug–drug interaction detection and AI-assisted triage to automate risk flagging across patient records
• JWT authentication with fine-grained RBAC; PostgreSQL with JSONB for semi-structured longitudinal patient histories

Python · NLP · Scikit-learn · PyTorch · NLTK · Deep Learning · BERT

End-to-end NLP system for dual-task classification on large-scale Amazon review data.

• Sentiment analysis (positive / negative / neutral) + deceptive review detection in a single pipeline
• Text preprocessing: tokenisation, stopword removal, lemmatisation, TF-IDF and word embedding feature extraction
• Trained and benchmarked Logistic Regression, SVM, LSTM, and BERT — evaluated on accuracy, F1-score, and ROC-AUC
• Ensemble methods and metadata feature engineering improved fake review detection accuracy

Python · Distributed Systems · Multiprocessing · Message Queues · Fault Tolerance

Task scheduler distributing compute workloads across parallel worker nodes with fault-tolerance and automatic retry on node failure.

• Implementing producer-consumer message queue patterns for inter-process communication and dynamic load balancing

MongoDB · Express.js · React · Node.js (MERN) · TailwindCSS

Full-stack food ordering platform with JWT-secured authentication, product listings, cart management, and end-to-end order processing.

• 🎨 ScoreMyResume — A web app that scores how well a resume matches a job description and returns actionable feedback
• ⚙️ Distributed Task Scheduler — fault-tolerant parallel computing system
• 🤖 AI/ML Coursework — Prime Complete AI/ML, Apna College

Open to AI engineering, backend, and ML internships / full-time roles · Bangladesh & Remote
sushmitah0199@gmail.com