Experience 👨🏻‍💻

Center for Intelligent Information Retrieval, UMass Amherst

Worked on further improving the perishable prediction algorithm by incorporating probability distributions and uncertainty estimation.
Laid the groundwork for personalized recommendations and marketing for the Chaldal platform.
Created Superset dashboards to monitor real-time performance of the ML models deployed in production, enabling quick identification and resolution of issues.
Ported the existing ML systems to our new Kubernetes based infrastructure.
Onboarded new team members and helped to get up to speed with the projects.

🏷️ Time Series Forecasting Recommender System Superset Kubernetes

Got promoted to Level 3 in record time of company’s history!
Developed a new Machine Learning driven perishable demand prediction algorithm that is a 70% improvement from its predecessor and is estimated to save 2000K BDT per month.
Integrated the new address search feature into the core Chaldal platform (website and app). Gives users a more accurate and faster address search experience in the checkout process.
Worked on creating a geospatial language model to help optimize delivery agents to deliver orders faster and more efficiently.

🏷️ Time Series Forecasting Information Retrieval

Led the research and development of a new address search feature. Improved the existing Lucene based search using fine-tuned LLM, retrieval augmented generation (RAG), and re-ranking. Improved the accuracy to over 60% from less than 10% after taking over the project. The project greatly reduced the need for humans in the loop for this essential process and has optimized delivery logistics.
Developed a reinforcement learning based in-house recommender system from the ground up for category based product recommendation.
Created an AB testing framework for the recommender system that is able to handle quick experiment iterations with no downtime! It is used to evaluate the performance of different recommendation models in production. Currently used to test on 200K+ customers.

🏷️ LLM Information Retrieval Retrieval Augmented Generation Reinforcement Learning RecSys

Remotely worked as a Research Intern at Dr. Min Xu’s lab at Carnegie Mellon University.
Worked on developing an annotation-efficient semi-supervised particle-picking framework for 3D object detection from macromolecular samples. The goal was to detect and classify particles from cryo-electron tomogramss having signal-to-noise ratio as low as 0.1, making it a very challenging problem. (Manuscript in preparation)

🏷️ 3D Object Detection Cryo-EM Semi-Supervised Learning Deep Learning

Developed a comprehensive online portal for advance payment application and tracking for the Directorate of Advisory, Extension and Research Services (DAERS), BUET.
Designed and implemented the backend from scratch to production, using Django Rest Framework, PostgreSQL, and Docker, according to the client’s custom need and workflow.

The system was launched in February 2023 for internal use at BUET.

🏷️ REST API Backend Development Django REST Framework PostgreSQL

Worked in Big Data Intelligence (BDI) Lab under Dr. Shubhra Kanti Karmaker, Assistant Professor, Auburn University, Alabama on Short Text Stream Clustering
Developed with a team a new software tool called One Pass Sentence Embedding Clustering (OPSEC) that is designed to efficiently cluster short text streams
Utilized a unique one-pass algorithm that calculates similarity scores based on sentence embeddings to cluster the short texts
Outperforms current state-of-the-art methods in terms of both accuracy and quality of clustering for short text streams

🏷️ Natural Language Processing Machine Learning Clustering

Worked with a team that developed a new sales forecasting model for Unilever Bangladesh
Applied data engineering techniques on raw sales data

🏷️ Data Engineering Forecasting