🎯 BI Analyst @ ShelfTrak | Applied AI Researcher

I am a Business Intelligence & AI professional with dual Master's degrees in AI & Data Science (Distinction) and International Business Management, combining deep technical expertise with real-world commercial impact. By day, I build advanced Power BI solutions for global travel retail clients across 50+ airports worldwide. Outside of work, I build and deploy AI projects spanning medical imaging, financial sentiment analysis, and time-series forecasting.

🧠 The "Why": I sit at the intersection of Retail Tech, Healthcare AI, and Financial Intelligence — domains where data-driven decisions directly affect real outcomes. I'm motivated by building AI systems that are not just accurate, but explainable, deployable, and genuinely useful.

💡 Technical Expertise:

• Languages: Python, SQL, DAX (Advanced), M (Power Query)
• Machine Learning & AI: CNNs, Vision Transformers (ViT, MaxViT, DeiT), XGBoost, LightGBM, CatBoost, LSTM, ARIMA, Transformers, Self-Supervised Learning
• Healthcare AI: Medical Image Classification, Clinical Decision Support, Proteomics Analysis, Grad-CAM Explainability
• NLP & LLMs: VADER Sentiment Analysis, LLM Integration, Agentic Workflows, Multi-Agent Systems
• MLOps & Deployment: Docker, FastAPI, Flask, React, Render, CI/CD Pipelines
• Agentic AI: Multi-Agent Systems, Phidata, Groq
• Financial Analysis: Market Sentiment Analysis, Time-Series Forecasting, Financial Data APIs (YFinance), Quantitative Modelling, Risk & Return Analysis
• BI & Visualisation: Power BI (Advanced), DAX, Power Query, Matplotlib, Seaborn

End-to-End MLOps & Medical AI | 🌐 Live Demo

A fully deployed web application for tumour classification, benchmarking 6 ML algorithms (Random Forest, XGBoost, CatBoost, LightGBM, SVM, Neural Network) with automatic best-model selection based on F1-score.

• Tech: XGBoost, FastAPI, Docker, Render
• Highlight: Real-time probability scoring with confidence levels and interactive tooltips explaining all medical terminology — built for non-technical clinical users

Academic Research | Medical Imaging | Deep Learning

Comprehensive benchmarking study comparing CNN, Vision Transformer, and hybrid architectures for automated glaucoma detection from 17,242 fundus images. The best-performing MaxViT-Tiny hybrid model achieved 99.76% accuracy.

• Tech: PyTorch, EfficientNetV2, DeiT, MaxViT, DINO SSL, Grad-CAM
• Methodology: Hyperparameter tuning via grid search, 5-fold cross-validation, ablation studies on preprocessing techniques
• Highlight: Grad-CAM visualisations confirm models focus on the clinically relevant optic disc region — validating real-world deployment readiness

MSc Dissertation | Bioinformatics & Clinical AI

A multi-method consensus framework for identifying biomarker candidates from platelet proteomics data, combining differential expression analysis, pathway enrichment, and unsupervised clustering to overcome small sample size limitations.

• Tech: Python, Bioinformatics pipelines
• Highlight: Novel consensus methodology designed to produce robust, statistically defensible biomarker candidates despite limited clinical sample availability

Full-Stack NLP Application | Real-Time Financial Intelligence

A full-stack financial sentiment dashboard aggregating breaking news from Bloomberg, CNBC, Reuters, and Yahoo Finance — analysing market sentiment with NLP and visualising live data for S&P 500, Gold, VIX, and Bitcoin.

• Tech: Python, Flask, React 18, VADER NLP, Docker, Render
• Highlight: Live RSS aggregation with Fear & Greed Index integration, interactive candlestick charts, and trend detection across 90-day history — no paid API key required for core functionality

Multi-Agent LLM System | Financial Research Automation

A Flask web application using a multi-agent AI system to deliver on-demand stock analysis — combining a Web Search Agent (DuckDuckGo) and a Finance Agent (YFinance) orchestrated via Groq's LLM to produce consolidated analyst ratings, price targets, and breaking news summaries for any stock ticker.

• Tech: Python, Flask, Phidata, Groq LLM, YFinanceTools, DuckDuckGo
• Highlight: Demonstrates real agentic AI architecture — two specialised agents with distinct tool access, coordinated to deliver structured financial intelligence through a responsive web UI

Time-Series Forecasting Benchmark

Systematic benchmarking of ARIMA, XGBoost, LSTM, and Transformer models for long-horizon energy consumption forecasting (2020–2040), with hyperparameter tuning and RMSE-based evaluation.

• Tech: PyTorch, XGBoost, Statsmodels, scikit-learn
• Highlight: LSTM outperformed all models under limited data conditions, demonstrating stronger generalisation than the Transformer — with scenario-based long-horizon forecasts generated via recursive strategy

• MSc Artificial Intelligence & Data Science (Distinction) | University of Hull (2023–2024)
• MSc International Business Management (Merit) | Sheffield Hallam University (2014–2015)
• BBA (Magna Cum Laude Distinction) | American International University — Bangladesh (2010–2013)

• 💼 LinkedIn: Shaon Biswas
• 📧 Email: biswas.shaon@gmail.com

💡 "Building AI that works in the real world — not just in notebooks."