Mehedi Hasan

AI/ML Engineer · Dhaka, Bangladesh · (+880) 1521323549 · abir.aust.102@gmail.com

An enthusiastic and results-driven AI/ML Engineer with extensive experience in developing and deploying cutting-edge machine learning solutions. Passionate about leveraging deep learning, computer vision, and LLMs to solve complex problems and drive business growth. Seeking a challenging role to contribute to innovative projects and advance my practical knowledge in the software industry.

Download CV

Experience

Senior AI Engineer

Next Solution Lab

Led the development of English DeepICR for document understanding (contracts/invoices), covering Text Detection, Layout Detection, Text Recognition, and Data Extraction. Implemented LoRA/PEFT-based LLM fine-tuning and built a RAG search system for Japanese legal documents. Optimized and managed deployment pipelines using a modular monolith architecture on AWS.

Key Technologies:
PyTorchTensorFlowHuggingFace TransformersLangChainDockerAWS (EC2, S3, ECR)ElasticsearchFastAPIOpenCVYOLOv7LayoutLM
Jun 2023 - Present

AI Engineer

Next Solution Lab

Built a robust Japanese Text-recognition pipeline with 5,000+ character coverage. Developed a rule-based engine for key-value extraction from documents, extracting 128 fields across 44 document clusters. Created a Proof-of-Concept for Smart Farming using camera-only cattle monitoring and developed pre-LLM Q&A systems for legal documents.

Key Technologies:
PythonOpenCVBERT/RoBERTaDetectron2YOLOv5StreamlitGoogle Vision APIRCNN+CTC
Jun 2021 - Jun 2023

Professional Projects

Libor DeepICR (English) - End-to-End Bank Agreement Processing

A production-grade document understanding pipeline for banking agreements, covering layout analysis, OCR, and entity extraction. Led the project from requirement analysis to client deployment, achieving high client satisfaction through improved model accuracy and reduced noise-induced errors.

Tech Stack:
PythonOpenCVYOLOv7DBNetMask R-CNNLayoutLM

DevOps & Delivery - English DeepICR Platform

Managed the end-to-end model deployment, monitoring, and release pipelines for a large-scale document understanding platform. Dockerized services and integrated FastAPI/Node.js backends with React clients, ensuring reliable uptime and smoother CI/CD operations on AWS.

Tech Stack:
PythonPyTorchDockerNode.jsReactAWS (ECS, EC2, S3)ElasticsearchCloudWatch

Libor Extraction Module - Rule-Based Key-Value Engine

Developed a deterministic OpenCV-based rule engine for key-value extraction, covering 128 fields across 44 document clusters of banking agreements. This delivered a high-precision baseline for downstream ML models, significantly improving data reliability.

Tech Stack:
PythonOpenCVREST API

Skills

Programming & Core Languages

PythonJavaScriptNode.jsC#JavaSQL

ML & Data Science

PyTorchTensorFlowKerasscikit-learnHuggingFace TransformersLangChainOpenCVPandasNumPySpacy

Cloud & DevOps

AWS (EC2, ECS, S3, Lambda, CloudWatch, CodeCommit)DockerCI/CDGitAgile/Scrum

Databases & Search

MySQLMSSQLElasticsearchOpenSearchChromaDB

Frameworks & Frontend

FastAPIStreamlitReactMaterial-UI

Education

Ahsanullah University of Science and Technology (AUST)

Bachelor of Science in Computer Science and Engineering

CGPA: 3.59

Apr 2016 - Jan 2021

SOS Hermann Gmeiner School and College

Higher Secondary Certificate

GPA: 5.00

2013 - 2015

Research Experience

Bengali Intent Classification with Generative Adversarial BERT

First Author · IEEE Xplore, 2023

Developed GAN-BnBERT, a generative adversarial BERT-based model for Bengali intent classification. Introduced the BNIntent30 dataset, demonstrating superior performance over Bi-LSTM and baseline BERT models.

Design of an Arrhythmia Classification Algorithm Using 2-D CNN

First Author · Undergraduate Thesis, AUST, 2021

Proposed a novel CNN-based arrhythmia detection algorithm from ECG signals, comparing performance with VGGNet for improved beat classification accuracy.

Awards & Accomplishments

  • 1st Runner-up, Intra AUST Programming Contest Fall-2017
  • Winner, Intra AUST Programming Contest Spring-2016 (Junior Section)
  • Qualified for Intra AUST Programming Contests (Fall-2016, Spring-2017, Spring-2018, Fall-2018)