|
AI/ML Engineer & Researcher applying explainable ML, deep learning, and multi-agent systems to two high-impact domains: Medical AI and AI for Power & Energy Systems. B.Sc. in Electrical & Electronic Engineering from Shahjalal University of Science and Technology, Bangladesh.
|
|
| Category | Technologies |
|---|---|
| Languages |      |
| ML & DL |       |
| LLMs & Agents |       |
| Deployment |      |
| # | Title | Venue | Link |
|---|---|---|---|
| 1 | Explainable ML for Multi-Class Power Quality Disturbance Classification Using SHAP | IEEE ICCIT 2025 |  |
| 2 | Quantifying Socioeconomic Status & Parent-Child Attachment on Adolescent Through ML | IEEE ICCIT 2025 | |
| 3 | Enhancing Wind Farm Operational Efficiency Through Power Output Classification | IEEE ICCIT 2025 | |
| 4 | Time Dependent Control of Voltage and Current via a Tunable Metasurface | IEEE ICCIT 2025 | |
| 5 | Robust Optical Pulling of Plasmonic, Chiral & Dielectric Nanorods via Hyperbolic Metamaterials | Elsevier Q1 (Under Review) |  |
| 6 | ViT-SkinNet: Vision Transformer for Monkeypox & Vesicular Skin Disorder Detection | In Preparation |  |
| 7 | TrustEHRAgent: Calibrated Multi-Agent Systems for Secure EHR Question Answering | In Preparation | |
| 8 | Explainable & Privacy-Preserving Federated Intelligence for Smart Grid Cyber Attack Detection | In Preparation | |
| Project | Description | Tech |
|---|---|---|
| LLM Hallucination Detection Pipeline | Production-grade factual accuracy evaluation on TruthfulQA (817 Qs, 38 categories) with interactive Flask/Plotly dashboard | 🔍 DeepEval · Flask · Plotly · OpenAI |
| RAG Evaluation Framework | Automated RAGAS-based evaluation suite (Faithfulness, Relevancy, Context Recall) with trade-off analysis dashboard | 📊 RAGAS · LangChain · ChromaDB · Flask |
| LLM-as-a-Judge System | GPT-4 judge evaluating weak model outputs on 4 rubrics (coherence, factuality, tone, safety) calibrated against human annotations | ⚖️ DeepEval G-Eval · Alpaca Eval · OpenAI |
| Prompt Regression Testing Pipeline | CI/CD quality gate that blocks prompt/model regressions using DeepEval + pytest + GitHub Actions on a 50-case golden dataset | 🛡️ DeepEval · pytest · GitHub Actions · OpenAI |
| Red-Teaming Framework for LLM Safety | Adversarial testing across 40+ vulnerability categories (jailbreaks, bias, PII, toxicity); reduced AdvBench violation rate from 56% → 0% | 🔴 DeepTeam · AdvBench · OpenAI · Python |
| AgroGPT | Bengali-inclusive multilingual voice-first AI advisory system with CI/CD | 🤖 FastAPI · React · Sarvam AI · Docker |
| MediPredict-X | Multi-disease risk prediction with LLM-based clinical insights | 🧠 FastAPI · Next.js · Llama 3 · Gemini |
| Medibot AI | RAG-powered clinical assistant with hallucination-minimized medical QA | 🏥 Groq · Pinecone · FastAPI |
| BengalVision ALPR | Real-time Bangla number plate detection at ~92% accuracy | 👁️ YOLOv11 · ONNX · OpenVINO |
| E-Passport AI Advisor | Multi-agent system for automated bilingual passport guidance | 🤝 CrewAI · Gemini · OpenAI |
| 100+ Language STEM Solver | Step-by-step Math/Physics/Chemistry solutions in 100+ languages | 🌐 FastAPI · Docker · Gemini 2.5 |
| AI Smart Exam Platform | Real-time automated grading for objective & subjective assessments | 📝 FastAPI · Docker · Groq Llama 3 |
Research-driven AI. Real-world impact.