Sadman Kabir Soumik.
Senior Data Scientist building AI systems that ship.
I architect and deploy large scale machine learning systems, including LLM agents, RAG pipelines, and recommender systems. Six years across research labs, AI startups, and platform companies. Currently at Optimizely.
Turning hard data problems into AI systems that move business metrics.
I’m a Senior Data Scientist focused on shipping production machine learning. My path into AI started with computer vision and NLP research, then widened into backend engineering, MLOps, and the messy work of putting models in front of real users.
Today I work on agentic AI, LLM memory components, RAG pipelines, and large scale recommender systems at Optimizely. Earlier in my career I shipped NLP and cloud automation at Venturas Ltd., and contributed to computer vision and NLP research at Chowa Giken Corporation in Japan.
About 70 percent of my day is reading, writing, and reviewing code. The rest goes into architecture, planning, and the hand to hand combat of getting cross functional teams aligned on what to build. I write about what I learn at soumik.blog.
- Currently
- Senior Data Scientist at Optimizely Inc.
LLM agents, RAG, recommender systems. - Previously
- AI Engineer at Venturas Ltd.
ML Engineer at Chowa Giken Corporation, Japan. - Education
- B.S. Computer Science and Engineering
North South University, Dhaka · 2015 to 2019 - Based in
- Dhaka, Bangladesh · open to remote
Stack
- Python
- TypeScript
- C/C++
- SQL
- Swift
- PyTorch
- Keras
- XGBoost
- scikit-learn
- HuggingFace
- LangChain
- OpenAI SDK
- Pandas
- NumPy
- OpenCV
- Tableau
- Matplotlib
- Kubeflow
- Apache Airflow
- Docker
- MLflow
- Kubernetes
- Terraform
- FastAPI
- Django
- Redis
- SQLAlchemy
- Alembic
- Celery
- BigQuery
- PostgreSQL
- Spanner
- Pinecone
- MySQL
- Elasticsearch
- Firebase
- GCP · Vertex AI
- GCP · GKE
- GCP · Cloud Run
- GCS
- Compute Engine
- AWS · SageMaker
- AWS · EC2
- AWS · S3
- Shell
- Unix/Linux
- LaTeX
- Scrapy
- Git
- Pytest
- Datadog
A working timeline of where I’ve shipped.
- Jan 2023 - Present
Optimizely is a U.S.-based multinational company that offers a digital experience platform as a Software-as-a-Service (SaaS) solution. The company provides A/B testing and multivariate testing tools, website personalization, feature toggling capabilities, web content management, and digital commerce solutions.
Things I've been doing here:
- Working on multi-purpose agentic AI systems, including architecting and developing LLM memory components that preserve user and chat context over extended periods.
- Architecting and implementing large-scale recommendation systems serving millions of daily user interactions.
- Developing sophisticated A/B testing frameworks for ML model evaluation and deployment.
- Building real-time model training pipelines for continuous learning from user interactions.
- Developing and maintaining end-to-end machine learning pipelines that train thousands of models daily.
- Creating conversational AI solutions for automated customer segmentation and analysis.
- Designing unified API services for multiple LLM providers to streamline enterprise-wide AI adoption.
- Implementing RAG-based systems to enhance AI applications with domain-specific knowledge.
- Doing cross-functional collaboration between data science and engineering teams for ML infrastructure development.
- Jun 2021 - Dec 2023
Venturas Ltd. is a Japanese startup offering human resource solutions to Japanese companies from Bangladesh. I joined the company as an AI Engineer in June 2021 and was promoted to Senior AI Engineer in January 2023. During my time there, I also took on the role of Tech Lead for a period.
Things I've done here:
- Created the overall structure and design of AI/ML systems, ensuring alignment with business goals, scalability, and performance requirements.
- Evaluated and chose appropriate technologies and frameworks that best fit the system's requirements.
- Provided technical leadership and guidance to development teams, facilitating knowledge sharing and best practices in AI/ML development.
- Ensured seamless integration of AI/ML components with existing systems and infrastructure, including cloud platforms like GCP/AWS.
- Conducted regular code reviews to ensure that the code meets the quality standards, follows the architectural guidelines, and adheres to best practices.
- Sep 2019 - Jun 2021
Chowa Giken Corporation is a Japanese artificial intelligence based R&D company that offers AI-based solutions for various industries headquartered in Hokkaido, Japan. I started my career here as a Machine Learning Engineer in 2019, immediately after completing my bachelor’s degree.
Things I've done here:
- Preprocessed data, including cleaning, normalization, and transformation for various machine learning tasks.
- Developed, fine-tuned, and optimized custom machine learning models for both NLP and computer vision applications.
- Analyzed and interpreted ML model outcomes to derive insights and assess performance.
- Deployed machine learning models into production using tools and platforms like Flask/Django and GCP.
- Conducted experiments with new techniques, updated models based on performance, and drove continuous improvement and innovation in machine learning systems.
Things I build and run outside of work.
bdtechjobs.com
A curated job board surfacing tech opportunities in Bangladesh. Designed, built, and maintained end to end as a solo project.
- Web app
- Solo build
- Live
Voice Writter
On-device voice to text dictation for macOS with automatic grammar correction, powered by Whisper and a local MLX language model. Works in any app; nothing leaves your Mac.
- Swift
- macOS
- On-device AI
- Open source
Reading Notes
A personal site for notes and highlights from the books I read. Built and maintained solo.
- Static site
- Solo build
- Live
PS Notes
A personal knowledge base of problem solving notes and quick references. Built and maintained solo.
- Static site
- Solo build
- Live
Notes from shipping AI systems in production.
I write about machine learning systems, infrastructure, and the parts of the job that don’t make it into papers. Full archive at soumik.blog.
Have an idea worth shipping? Let’s talk.
I’m always open to a thoughtful conversation about AI systems, recommender engines, or interesting engineering work. Drop a line and I’ll get back to you soon.