Khandker Aftarul Islam

Software Engineer (Backend & DevOps)

• Developed a video and image processing pipeline using Gemini to ground and analyze ∼100,000 daily social media assets, enabling precise content referencing
• Built a high-performance analytics pipeline on Google Cloud Platform (GCP) using Pub/Sub, Cloud Tasks, Cloud Scheduler, Cloud Run, and Artifact Registry, processing ∼5 Million social media contents daily to track client brand sentiment and product insights
• Implemented an in-house AI transcription service with faster-whisper model, replacing OpenAI's API, reducing costs by 75% and improving speed and throughput
• Automated CI/CD pipelines with GitHub Actions and Infrastructure as Code (IaC) using Terraform to manage VMs, containers, and queues, caching pre-built steps to increase deployment speed by 37%

Software Engineer II

• Designed and implemented Emoji Reactions feature using React, Node.js, MongoDB, and WebSocket, boosting user engagement by 65% through real-time content interactions
• Engineered All-Day Events feature for Calendar module with Node.js, MongoDB, and React, resolving timezone display issues and streamlining event management for global teams
• Developed Recycle Bin feature with soft-delete functionality and 60-day auto-cleanup system, reducing manual asset restoration requests by 90% in collaboration with frontend team

AI & Software Engineer

• Built a Keyword Spotting Model for British American Tobacco using TensorFlow, achieving 87% accuracy for marketing campaign audio analysis, surpassing competing solutions after preprocessing with Spectral Gating
• Created a Face-Anti-Spoofing Model for sheba.xyz using Keras, achieving 94% accuracy on video test datasets, overcoming real-time video processing challenges
• Deployed a Kubernetes cluster on bare-metal infrastructure with Kubeadm to manage AI models, analytics, backend, frontend, RabbitMQ, and Postgres, ensuring high uptime and scalability via Horizontal Pod Autoscaling (HPA)

Software Engineer

• Developed smart contracts and backend for Immunodex, a Hyperledger Fabric-based Vaccination Certificate Platform, enhancing security for educational institutions in Hong Kong
• Designed and built frontend for StackEd, a verifiable certificate-sharing platform using Hyperledger Fabric, enabling secure credential issuance

Mathematical Bridge Between Epidemiological and Molecular Data on Cancer and Beyond

Role: Author

Developed a mathematical model to integrate epidemiological and molecular data for cancer research, analyzing 21 cancer categories and 14,087 cases

Online Detection of Attentiveness of Students with Special Needs

Role: First Author

Developed a deep learning-based model to predict attentiveness levels of students with Autism Spectrum Disorder (ASD) in online classes, using real-time video data to classify behaviors like gaze and movement

Dehydration Scan: AI-Assisted Smartphone-Based System for Early Detection of Dehydration

Role: Author

Designed an AI-assisted smartphone app to detect dehydration by analyzing facial landmarks, achieving 76.1% accuracy in identifying dehydration levels

3D-FFS: Faster 3D object detection with Focused Frustum Search in sensor fusion based networks

Role: Co-First Author

Proposed 3D-FFS, a sensor fusion-based 3D object detection method, improving inference times by up to 62.80% and reducing memory usage by 58.33%