Md. Motahar Mahtab
Greetings!
I am Md. Motahar Mahtab, an AI Engineer with 2.5 years of industry experience in building state-of-the-art ML models with a solid academic background and hands-on experience in the tech sector. I graduated from BRAC University with a CGPA of 3.99 in Computer Science & Engineering. Currently, I am currently working at Delineate Inc., where I have worked on medical data extraction from Quantitive Systems Pharmacology papers using GPT4o in multi-agent pipelines. Previously, I was with Giga Tech Ltd., developing systems for various Bangla NLP tasks, creating REST APIs using FastAPI, optimizing ML models with Nvidia TensorRT, and deploying these models on the Nvidia Triton Inference Server for efficient inference.
My technical skill set includes ML libraries (like PyTorch, Huggingface, LangChain), web frameworks (Flask, Django, FastAPI, Streamlit), and tools for ML optimization and deployment (such as Triton, Dask, and MLflow).
Experience
- Parallel background processing using Celery and Webhooks for faster Table Extraction using GPT4o.
- Pipeline enabling extraction of QSP clinical trial data from clnical papers and tables.
- Created new state-of-the-art systems for a plethora of Bangla NLP tasks e.g. Named Entity Recognition (NER), Parts of Speech (POS), Lemmatization, Question Answering, Coreference Resolution and Emotion recognition. Performed R&D on increasing performance beyond the current state-of-the-art to achieve 90% KPI on ML modules. Two such systems Bangla Lemmatization and Emotion recognition are publicly available at https://github.com/eblict-gigatech/BanLemma and https://sentiment.bangla.gov.bd respectively.
- Created GPT4o inference pipeline for Bangla NER and Coreference Resolution systems using ReAct prompting method achieving comparable performance against finetuned systems.
- Created pipeline for Natural Language generation (NLG) in Bangla for both encoder models like BERT and auto-regressive models like GPT2. Analyzed and overcame common issues like repetitive text generation, and unmeaningful word generation in NLG for Bangla.
- The Question Answering (QA) module establishes new state-of-the-art results on Bangla datasets including SQuAD-bn (translated from the SQuAD-2.0 and TyDI-QA English QA datasets) by a modified loss function to balance performance among null and non-null questions.
- The NER classification module establishes new state-of-the-art results on Bangla NER datasets by a hierarchical majority voting mechanism among external contexts retrieved from a Knowledge Base.
- Created data augmentation pipeline to handle the class imbalance problem in sequence tagging tasks. Formulated a general test set creation guidelines for unbiased classification performance calculation.
- Optimized deployment of LLMs using Optimum (for ONNX conversion) and Nvidia TensorRT(TRT) format for further optimization. Used PyTorch Profiler to identify inference bottlenecks. Used Nvidia Triton Inference Server (TIS) as the default ML inference server for concurrent request serving and scheduling, batch inference and response caching in MongoDB. Used Locust for load testing and pytorch profiler to reduce bottlenecks.
- Created REST APIs using FastAPI for hosting ML inference endpoints. Used MongoDB for response caching in NVIDIA Triton.
- Used Qdrant vector DB for fast semantic searching, Dask to analyze and query big dataframes, DVC for dataset versioning and MLflow for model, artifact and experiment versioning.
- Used Qdrant vector DB for fast semantic searching, Dask to analyze and query big dataframes, DVC for dataset versioning and MLflow for model, artifact and experiment versioning.
- Pretrained a HuBERT model on Bangla ASR dataset for joint task of speech and speaker recognition pipeline using SpeechBrain.
- Assisted in enriching existing open source Bangla ASR datasets by adding more scripted audio and correcting existing annotation
- Helped students with different coding assignments and helped teachers in checking scripts.
- Assisted students in conducting research in various fields and submitting papers to conferences.
- Assisted Teachers in lab classes and helped students with different course materials during consultation hour.
Education
Publications
- State-of-the-art Bangla Rule Based Lemmatizer which proposes a novel iterative suffix stripping approach based on the part-of-speech tag of a word.Shows superior performance than all previously published Bangla lemmatization methods on existing datasets.
- Paper BanLemma A Word Formation Dependent Rule Based Lemmatizer
- Code https://github.com/eblict-gigatech/BanLemma
- Supervisor Dr. Nabeel Mohammed, Dr. Ruhul Amin
- First Bangla Clickbait News Article Dataset containing 15,056 data instances. Investigated various semi-supervised learning methods and compared them with supervised learning methods to prove the former’s superiority.
- Paper BanglaBait: Semi-Supervised Adversarial Approach for Clickbait Detection on Bangla Clickbait Dataset
- Code https://github.com/mdmotaharmahtab/BanglaBait
- Supervisor Dr. Farig Sadeque
- Trained state-of-the-art Transformer networks in adversarial fashion using Generative Adversarial Network (GAN) to achieve superior performance when labelled dataset size is too small.First Bangla Paper to investigate the application of GAN-BERT on Bangla text classification tasks.
- Paper A GAN-BERT Based Approach for Bengali Text Classification with a Few Labeled Examples
- Supervisor Mr. Annajiat Alim Rasel
Projects
Bangla Clickbait Detector (Pytorch, Streamlit, Node.js)
Demo app created as a part of research work on Bangla Clickbait Detection using GAN-Transformers. It takes a Bangla article title as input and outputs whether the title is a clickbait or non-clickbait along with the prediction probability score. GAN-Transformers is a Transformer network trained in a generative adversarial training framework.
Bangla Article Headline Categorizer App (Pytorch, Streamlit, Node.js)
Can categorize Bangla article headlines into eight different categories - Economy, Education, Entertainment, Politics, International, Sports, National, and Science & Technology. Models used State-of-the-art Bangla ELECTRA model, Dataset used Patrika Dataset - contains ~400k Bangla news articles from prominent Bangla news sites.
EBRAC - Online Learning App (Django, Bootstrap, Node.js)
A comprehensive online education platform where instructors can create different courses, upload course content, enrol students, see students’ marks, prepare questions, take quizzes etc. Students can enrol in courses, view course contents, participate in exams and see results.
Veggie (Django, Bootstrap, Node.js)
This web app allows users to view different vegetarian recipes, and see their total calories, nutrients like protein, carbohydrate, fat and their ingredients. Users can create their own vegetarian recipes by mixing different ingredients available on the web app. They can also see the total nutrients and calories of their created recipe
Skills
90%
90%
90%
90%
95%
85%
80%
75%
90%
95%
95%
90%
80%
80%
80%
90%
90%
70%
70%
70%
90%
85%
80%
70%
Accomplish
ments
ments
Certifications
Open Source
Contributions
Contributions