Engineering
Production-Grade Intelligent Systems
The Algorithm Behind Me
A Full-Stack AI/ML Engineer with a pure mathematics foundation, operating at the intersection of deep learning research and high-throughput software architecture.
I eliminate the production gap for AI systems. I transform raw research notebooks into containerized, low-latency APIs capable of running real-time inference at scale.
By combining state-of-the-art vision models (ViTs, Diffusion, YOLO architectures) with robust backend engines (FastAPI, Redis caching, message queues) and automated infrastructure.
Distributed training optimization, Latent Diffusion Models, Vision-Language coordination, and hyper-contextual e-commerce reasoning engines.
Experience & Education
Building products, mentoring students, and studying the math behind it all.
Founder & Lead Architect — Memora AI
Designed scalable distributed cluster sorting for massive event photo grouping — facial embedding extraction, similarity-based clustering, and one-click personalized retrieval at scale.
AI Engineer — Faculty Finder Outreach Automation
Implemented a sub-second semantic RAG pipeline over institutional databases, combining dense vector search with BM25 hybrid ranking for targeted academic outreach.
Computer Vision Engineer — Skin AI BD
Optimized multi-class medical image classification for edge inference deployment — data-efficient transformers, class-imbalance handling, and INT8 quantization.
Deep Learning Mentor — SUST Data Science Club
Directed project-based curriculums on Vision Transformers and object tracking, mentoring 50+ students through hands-on CNN and applied deep learning sessions.
B.Sc. in Mathematics — Shahjalal University of Science & Technology (SUST)
Sylhet, Bangladesh · Focus: Linear Algebra, Calculus, Probability, Differential Geometry, Numerical Analysis.
Technical Arsenal
Click on any skill to activate neural pathways.
Agentic AI & Research
Capability Matrix
Flagship Systems
Production architectures — from problem statement to pipeline, model tier, and measured performance.
Memora AI
Distributed Image Clustering Engine
Manual event photo sorting scales quadratically O(N²) in friction for users. Systems need automated, privacy-first, identity-grouped sorting at scale.
PropVision AI
Satellite & Geospatial Intelligence Platform
Raw satellite imagery lacks actionable property analysis, environmental risk scoring, and structured report automation for enterprise land valuation.
Faculty Finder
Hyper-Contextual Outreach Engine
Academic research outreach is bottlenecked by fragmented institutional profiles and manual email scraping routines.
Skin AI BD
Advanced Medical Image Diagnostics
Dermatological tracking suffers from high visual variance, low training set availability, and steep computational resource constraints for deployment on localized mobile hardware.
GuardianNet AI
Real-Time Urban Hazard & Safety Vector Tracking
Video-based danger tracking routinely trips on false positives and struggles with processing multiple multi-stream video setups concurrently.
Photo2Math
Optical Mathematical Parser & Renderer
Converting unstructured handwritten mathematical expressions into interactive graphs requires handling varying spatial stroke arrangements.
More Builds
GuardianNet AI
Built at the EWU Hackathon: a real-time hazard-detection safety dashboard triggering automated alerts on object detection, integrating Vision Transformers and pre-trained vision models for fast, high-fidelity classification. Designed a three-tier response protocol with medical logistics integration.
Nexus Style
A hyper-contextual e-commerce concierge chatbot built with LangChain and LangGraph that adapts recommendations to localized weather, geolocation, and user attributes. Deployed as an agentic AI commerce engine on FastAPI.
Agentic AI using LangGraph
Exploratory agentic AI workflows built with LangGraph, chaining reasoning and tool-use steps into stateful multi-agent graphs.
Real-Time Subject Blur
After every wedding, tour, or event, photos scatter across 10-20 phones and a shared Drive folder turns into 1000+ unsorted images. Built to solve the adjacent privacy problem: detects people in a live video stream and applies real-time blurring for privacy protection, anonymizing research footage, and surveillance use cases — hands-on experience in detection, segmentation, and video pipelines.
Memora AI (Capstone)
After every wedding, tour, or event, photos are scattered across 10-20 phones — a WhatsApp group or Drive folder turns into 1000+ uploaded photos, and finding your own becomes a headache. Memora AI fixes this: create one shared link, everyone uploads once, and AI automatically detects faces and clusters photos so each guest gets only their own — no manual searching, no signup, one-click personalized downloads.
Photo2Math (Capstone)
Upload any photo (best as line drawing) and let AI instantly generate mathematical equations that recreate its shapes and contours. These equations render live as interactive, explorable graphs turning images into stunning mathematical art — a unique fusion of computer vision and math visualization.
Project Website (This)
My personal portfolio website built with Next.js, TypeScript, and Tailwind CSS.
Email Spam Classifier
A machine learning model using techniques like Naive Bayes for spam detection.
Laptop Price Prediction
End-to-end ML pipeline that predicts laptop prices from specifications: data preprocessing and feature engineering, model tuning across Linear, Ridge, Lasso, KNN, Decision Tree, and Random Forest with GridSearchCV hyperparameter search, served through an interactive Streamlit app and deployed on Render.
WhatsApp Chat Analyzer
A data analysis tool to analyze WhatsApp chat data providing insights.
Movie Recommendation System
Collaborative & content-based recommender. Deployed as an API.
IPL Winning Probability
Predicting the winning probability of an IPL team mid-match.
Cat vs. Dog Image Classifier
Deep learning model using CNNs to classify images.
Face Expression Recognition
Real-time face expression recognition system.
Parking Spot Detection
Automated system to detect available parking spots.
Air Canvas
Gesture-based virtual drawing application — draw in the air using just your fingers, tracked and rendered live with OpenCV and MediaPipe hand landmarks.
Real-time Object Alert
Detects objects in a live video stream using YOLOv11 + OpenCV and automatically sends a WhatsApp alert message the moment a detection occurs — built for security surveillance, industrial safety, and smart monitoring use cases.
Kaggle Achievements
Competitive machine learning, ranked against the world.
RoadVision
Detecting and classifying vehicles in Bangladesh's highway surveillance imagery. Community competition · 124 teams.
PoliMemeDecode
Deep-learning classification of political context from multimodal meme data.
Shobdotori — Bangla Dialect Transcription
NLP normalization models for diverse regional Bangla dialects.
MemeDecode — Humor Classification
Multi-label architectures categorizing abstract web content.
AI-FICATION — Bengali STEM Translation
Neural translation of technical Bengali material into English STEM content.
Hackathon Finalist
Building under pressure, on stage, against the clock.
The Infinity AI BuildFest 2026
National AI hackathon bringing together 3,360+ participants and 612 teams from universities across Bangladesh, building production-ready AI solutions across Education, Healthcare, Marketing, E-Commerce, and InfoTech tracks.
Techathon Nationals & Rover Summit 2026
Advanced to the Final Round of the Hackathon Segment of Techathon Nationals and Rover Summit 2026, organized by the Robotics Society at Islamic University of Technology (IUT).
Research & Reading Radar
Active domains of study, and how core tooling is classified by maturity.
Computer Vision
- ▹Vision Transformers (ViT, DeiT, Swin)
- ▹Self-Supervised Learning
- ▹Masked Autoencoders
Generative Systems
- ▹Diffusion State Spaces
- ▹Latent Diffusion Optimizations
- ▹Multi-Modal Cross-Attention
Reasoning Models
- ▹Vision-Language Coordination
- ▹Chain-of-Thought in Multi-Modal Systems
Live Technical Tech Radar
MLOps Deployment Registry
Operational verification for systems currently running in production.
| Endpoint | Platform | Status | Metrics | Links |
|---|---|---|---|---|
| api.memora.ai/v1 | Railway / Custom Docker | Live | 99.98% Uptime (UptimeRobot) | |
| propvision.ml/inference | Render / Containerized | Active | Avg Response: 180ms | |
| nexus-stylist.engine | Vercel Edge Runtime | Live | Edge Cache Hit Rate: 92% |
Git Analytics
Live from the GitHub API — language mix, spotlight repos, and recent activity.
Language Distribution
Core Repositories
- SUST_Preli★ 2⑂ 0Python
- ICT_Fest_Hackathon_Preliminary★ 1⑂ 0Python
- IUT-Hackathon★ 1⑂ 0Python
- ticket_api★ 1⑂ 0Python
Recent Activity
- Agentic-Ecommercetoday
- ICT_Fest_Hackathon_Preliminary16 days ago
- IUT-Hackathon21 days ago
- SUST_Preli29 days ago
- ticket_api1mo ago
Ready to Collaborate?
I'm currently available for freelance work and full-time positions.
Average response window for technical inquiries: <12 Hours
Initialize Handshake