Software Engineer specialized in Back-end development with Python. 3 years of professional experience in building robust and scalable software. Expertise in FastAPI, MongoDB, ElasticSearch, GCP, LangChain and NLP. Experience in recommender systems, data pipelines, machine learning and generative AI.
See my complete CV on Google Drive
(Formerly CONCURED Limited) Mar 2024 – Present
Overview
ELELEM AI is an R&D venture exploring the potential of using State-of-the-Art AI technology to provide solutions for businesses. We primarily leverage the power of Large Language Models (LLM) to build automated agents. At ELELEM AI, we research the latest advances in the field of AI and LLM, propose new problems to be solved, and find their solutions. As a continuation of CONCURED LTD. with a renewed vision, we maintain many of its services, such as the Recommendations API.
- Responsibilities:
- Research and Development: Automated agents using LLM
- Researching the state of the art AI technologies
- Exploring the potential of LLM in order to build automated solutions
- Prototyping agents with LLM frameworks like LangChain, LangGraph
- Recommender System and API maintenance
- Maintaining the REST API services providing recommendations (FastAPI)
- Maintaining Backend processes, data pipelines for the recommender system
- Maintaining the overall cloud native system (GCP)
- Research and Development: Automated agents using LLM
Overview
EBLICT stands for Enhancement of Bengali Language in ICT through the Research & Development, a division of the Ministry of Posts, Telecommunications and IT of the Government of Bangladesh. Under EBLICT, I have been working as a Software Engineer responsible for building the system from a Backend perspective. This involves designing the system architecture, overseeing development of Backend and Frontend APIs. This is a project that involves multiple teams dedicated to developing and implementing the machine learning model, building websites and apps with Frontend development, managing different platforms for managing data and operation. My role here is to take the lead in communication and collaboration between all the teams.
- Responsibilities:
- Backend Lead
- System Architecture Design and Implementation for the "Virtual Private Assistant (SD-20)" project
- Backend Lead
Dec 2023 – Feb 2024
Overview
At Hiperdyne Corporation, I worked as an AI Engineer in both R&D and application development for 3 months. I researched Jailbreak in LLM and developed Computer vision applications. I did not renew my contract after the 3-month probation period, as the other company (Concured) I was committed to offered me a new and better contract with promotion to a more managerial role that requires more productivity and commitment. However, my time with the Japanese was really helpful in terms of interacting with people from a very different social and cultural background. It provided me useful insights and helped me grow as an Engineer in a multicultural setup.
- Responsibilities:
- Computer Vision Project: Face Privacy of Images & Videos
- Using different face detection models to hide the identity of selective individuals in image/video
- NLP/LLM Project: Research on Jailbreaking of Large Language Models e.g. ChatGPT
- Collected and studied research papers related to the security of Large Language Models in terms of its alignment to AI ethics. Explored various works done so far from both an attacking and defending perspective. A clear understanding of "jailbreaking" is crucial in securely deploying any public-facing service leveraging LLM or AI for production.
- Computer Vision Project: Face Privacy of Images & Videos
(Rebranded to ELELEM AI from Dec'23) Feb 2022 – Feb 2024
Overview
The primary product/service at CONCURED is the Recommendations System and its associated API. The recommendations provided are links to content/articles web pages generated for other content/articles pages. During my tenure at the company in various roles, I developed models for generating recommendations, data pipelines for the pre-processing steps of data crawling/scraping and data wrangling (finding keywords, text embeddings), served recommendations as REST API, and performed DevOps tasks to deploy all of these processes on GCP.
- Responsibilities:
- Recommender System and API maintenance
- Contribute to the REST API providing recommendations (FastAPI)
- Data analysis and visualization to evaluate the recommendations
- Leadership and supervising junior members
- Maintaining Backend processes, data pipelines for the recommender system
- Performance Evaluation of the recommendation models
- Stored and analyzed user information (requests and response) with Google's BigQuery
- Developed methods to evaluate and compare models based on various performance metrics
- Collaborated with team members to build Looker dashboards and write interactive Jupyter notebooks for data visualization
- Modularizing services and encapsulating them as microservices and API
- Divide complex data pipelines and process entanglement on GCP into smaller blocks of microservices
- Use micro-services as stand-alone APIs and refactor backend processes to consume them
- Refactor system architecture after careful design planning
- Expose micro-services as 3rd party services to potential users for usage in production
- Recommender System and API maintenance
- Projects:
- Development of Vector based recommendation model using text embeddings to generate user Interest Profile
- Description:
- Development of a new model using Natural Language Processing (NLP) providing better and more accurate recommendations. The model was deployed on Cloud Run (GCP). The model works by clustering documents into an embedding space, then finding the right cluster for a user based on their profile.
- Prototyping the model for testing, beta release for A/B testing.
- Enhancing the model latency by improving the design architecture with optimal technology choices.
- Technologies used for generating text embeddings include the GloVe model trained on Wikipedia (2014) with Gensim and the PMMBv2 model from HuggingFace with SentenceTransformers
- Outcome:
- Successfully built a new recommendation model using free and open-source tools ensuring excellent accuracy and speed. The addition of the new model provided depth and versatility of recommendation model stack.
- Description:
- Development of Vector based recommendation model using text embeddings to generate user Interest Profile
- Responsibilities:
- Maintaining Backend processes and the data pipeline
- Developing new Backend processes
- QA, testing and monitoring all backend processes
- Projects:
- Keyword Extraction and labeling with WikiData using YAKE algorithm and text embeddings (NLP)
Crop and bound-box interesting regions of an image (smart cropping) from saliency maps generated with SAM-LSTM-RESNET model (details here)
- Languages: Python (Expert with 3+ years of experience), JavaScript: React/NextJS (intermediate)
- Databases: NoSQL: MongoDB, ElasticSearch, Redis; SQL: BigQuery
- Backend Frameworks: FastAPI, Django, Flask
- Agentic Frameworks: LangChain, LangGraph
- ML Frameworks: TensorFlow, PyTorch, ONNX (NVIDIA CUDA as backend)
- Container Orchestration: Docker, Kubernetes
- CI/CD: GitHub actions, Cloud Build (GCP)
- Parallelism: AsyncIO
- Data Science Stack: Pandas, NumPy, SciPy, Scikit-Learn, NLTK
- OS: Linux, Windows
- Cloud Infrastructure: GCP (Expert with 2.5 years of experience), AWS (Personal Web Development projects)
- DevOps | System Architecture | Team Lead
Mail me at biis.saadi@gmail.com
LinkedIn: https://www.linkedin.com/in/sheik-sadi-rohmotullah/