Hello, I'm Yafi.

Software Engineer · Bioinformatics Researcher

Dhaka, Bangladesh · open to relocation

Portrait of Mashrur Ahmed Yafi

About Me

I am a software engineer with + years of experience. I lead an AI Assistant team at IQVIA, building agentic applications on LangGraph and LangChain for global pharmaceutical clients, from architecture through production operation. I am also the first author of scGIST (Genome Biology, 2024), a deep-learning method for gene panel design in spatial transcriptomics that began as my undergraduate thesis.

Experience

IQVIA

Software Development Engineer III (May 2026 - Present) · II (Oct 2024 - Apr 2026) · I (Sep 2023 - Sep 2024)

Dhaka, Bangladesh

  • Lead an AI Assistant team, owning architecture, delivery, and production operation of agentic applications for global pharmaceutical clients.
  • Build AI agents with LangGraph and LangChain, using tool calling, state management, and human-in-the-loop approval.
  • Developed MCP servers exposing data and analytics tools to agents, letting users run analysis conversationally.
  • Implemented retrieval-augmented generation over enterprise datasets with AWS Bedrock, GraphRAG, and Snowflake Cortex Search, grounding responses in governed, access-controlled data.
  • Redesigned the agent graph, cutting end-to-end latency by 15% and token usage by 10%.
  • Created a monorepo and introduced spec-driven development, reducing development time by 40%.
  • Deploy and operate agent servers on Kubernetes (AWS EKS) with GitLab CI/CD, and debug agent traces in LangSmith.
  • Previously led a .NET backend team, designing microservices and REST APIs on .NET Core and SQL Server with RabbitMQ for asynchronous messaging.
  • Reduced export-service memory usage by 85% and improved several API response times by up to 10x.

Enosis Solutions

Software Engineer II (May 2023 - August 2023) · I (May 2022 - May 2023)

Dhaka, Bangladesh

  • Developed a CRM web application in .NET Core and Angular for advertising campaigns across broadcast and digital channels, following the Kanban methodology.
  • Automated performance metrics retrieval from the Google Ad Manager API using timer-triggered Azure Functions.
  • Optimized complex queries in Microsoft SQL Server, significantly improving data retrieval performance.
  • Created automated statistics reports with scheduled email delivery to subscribed clients.
  • Migrated a legacy .NET MVC project to .NET Core.

Research & Publications

Research Interest

I am interested in applying artificial intelligence in bioinformatics, with a focus on single-cell RNA sequencing, spatial transcriptomics, and visualizing high-dimensional data. I am also interested in developing and optimizing machine learning algorithms for these problems.

Areas of interest:

  • Computational Biology
  • Single-Cell & Spatial Transcriptomics
  • Bioinformatics
  • Machine Learning

scGIST: Gene Panel Design for Spatial Transcriptomics with Prioritized Gene Sets

First author · Published in Genome Biology in February 2024

Yafi, M., Hisham, M., Grisanti, F. et al. scGIST: gene panel design for spatial transcriptomics with prioritized gene sets. Genome Biology 25, 57 (2024).

The initial goal of the project was to modify a simple linear programming algorithm to include prioritized genes, which we achieved in a short period of time. After that, we challenged ourselves to achieve even better results. Through several iterations and experiments, we developed scGIST.

  • Developed a deep-learning tool for designing single-cell spatial transcriptomics gene panels.
  • Created a custom loss function for constrained feature selection, optimizing both cell type detection and inclusion of prioritized genes.
  • Achieved high accuracy in cell type identification across multiple datasets, outperforming existing methods, especially for small panel sizes.
  • Enabled researchers to systematically prioritize genes of interest, including easy-to-target genes and receptor-ligand complexes, without sacrificing accuracy.

Projects

JSON Escape / Unescape

  • A VS Code and Cursor extension for escaping and unescaping JSON strings in place.
  • Built with TypeScript and Node.js and published to the VS Code Marketplace and Open VSX.
  • Over 2,200 downloads.
TypeScript Node.js VS Code API

Aero Color

  • A machine learning model to colorize any given grayscale image.
  • Created a U-Net model using Keras to train on 25,000 images.
  • Used OpenCV to pre-process and post-process the images.
Python Keras Tensorflow

Skills

Languages

Python R C# JavaScript TypeScript C++ Java SQL

AI & Agents

LangGraph LangChain MCP RAG GraphRAG LangSmith Snowflake Cortex Vector Databases

ML & Bioinformatics

Keras TensorFlow Pandas NumPy Seurat Scanpy anndata

Web & Data

.NET Core Angular React Next.js PostgreSQL SQL Server MongoDB Snowflake

Cloud & DevOps

AWS Azure Docker Kubernetes GitLab CI/CD RabbitMQ Git

Education

Bachelor of Science in Computer Science & Engineering

Bangladesh University of Engineering and Technology

February 2017 - May 2022

CGPA: 3.73/4.00

Notable Courses: Bioinformatics, Machine Learning, Operating System, Software Engineering, Compiler, Computer Security, Database, Computer Networks, Computer Architecture, Artificial Intelligence, Microprocessors and Microcontrollers, Data Structures and Algorithms, Object Oriented Programming Language.

Awards & Achievements

Test Scores

Get In Touch

Let's connect and discuss opportunities