Masum Billal
Verified Expert in Engineering
Software Developer
Kuala Lumpur Federal Territory of Kuala Lumpur, Malaysia
Toptal member since July 14, 2022
Masum is a software architect and AI engineering lead with 8+ years of experience designing high-performance back-end systems, distributed microservices, and enterprise-grade data pipelines. He specializes in bridging the gap between advanced machine learning and GenAI orchestration and robust, production-ready software engineering.
Portfolio
Experience
- Back-end Development - 8 years
- Data Engineering - 8 years
- Machine Learning - 8 years
- Microservices - 8 years
- Python - 8 years
- Software Architecture - 7 years
- Generative Artificial Intelligence (GenAI) - 6 years
- FastAPI - 5 years
Preferred Environment
Linux, Python, Back-end Development, Machine Learning, Generative Artificial Intelligence (GenAI), Software Architecture
The most amazing...
...architecture I've designed converts high-volume unstructured data into deterministic, high-quality, fully traceable intelligence at scale.
Work Experience
Python Expert
Farm To People LLC
- Rearchitected a legacy codebase with faulty implementations to improve response time from minutes to mere seconds.
- Detected and fixed all SQL injection-prone error points within the codebase.
- Established the framework for proper testing for the whole team, along with coding standards and CI/CD.
Engineering Tech Lead
Ur AI Co., Ltd.
- Architected end-to-end multimodal document intelligence microservices integrating ingestion, retrieval, reasoning, and inference orchestration using FastAPI, Redis, PostgreSQL, Alembic, LangChain, and AWS.
- Designed a unified, distributed, event-driven compute orchestration layer with large language model (LLM) and GPU worker pools, enforcing global RPM/TPM limits across all services.
- Developed an ingestion pipeline for the asynchronous processing of PDFs, Excel files, HTML, images, and JSON, with page- and sheet-level normalization and OCR, processing more than 100 documents in minutes.
- Built a hybrid retrieval-to-reasoning pipeline: dense (pgvector and ANN), BM25, and knowledge graph retrieval, RRF ranking, cross-encoder reranking, and token budget fitted LLM-based final selection.
- Introduced a structured tabular reasoning layer: SQL-based aggregation for millions of rows, replacing direct retrieval bottlenecks.
- Ensured full traceability of all model outputs by grounding every generated claim in source-level evidence (page, table, and sheet).
- Built a production chatbot for 500+ documents with 20- to 30-second response latency.
- Implemented observability using Datadog and OpenTelemetry, as well as React and a Bun front end.
Data Scientist
SoilSerdem LLC
- Improved the accuracy of a geospatial machine learning training process from 43% to 96%+.
- Developed an alternative to QGIS Grass plugins supporting features like line generalization algorithms, such as smoothing and simplification, without relying on Grass.
- Modified the training process to include a buffer and predict the values for the removed points based on features such as remote sensing data, yield, soil map data, and DTA.
Python Engineer (via Toptal)
Toptal
- Leveraged asynchronous FastAPI and SQLAlchemy, improving LLM API performance by over 400% and reducing latency.
- Designed a universal API framework for LLM integration, cutting vendor onboarding time by over 80%.
- Collaborated on 8+ cross-functional AI projects, ensuring scalable architecture and seamless deployment.
- Mentored team members on best practices and improved CI/CD pipelines, leading to fewer bugs and faster development.
Senior Python Developer | Computer Vision Engineer
Advanced Mobility Analytics Group
- Built real-time data pipelines for traffic analytics with 20% lower resource consumption with YOLO, PyTorch, ONNX, and TensorRT.
- Developed high-performance near-real-time data pipelines for traffic monitoring using YOLO v7 and OpenCV.
- Refactored legacy codebase into a well-maintained modern repository with high code coverage and best practices.
Expert Python Developer (via Toptal)
Woven
- Developed and optimized automated testing applications for mapping platforms, improving CI/CD performance and reducing testing costs.
- Improved data processing, validation performance, and CI execution time by around 90% using optimization techniques such as vectorization.
- Defined best practices and standards for improving code quality and better maintenance.
Senior Data Scientist | ML Engineer
iXora Solution
- Engineered event-driven, real-time, and scheduled data pipelines in GCP, Django, Flask, MongoDB, and AWS.
- Optimized ETL memory usage by over 95%, enabling large-scale data handling.
- Developed artificial intelligence solutions using Django, NLP, and machine learning.
- Saved a client thousands of dollars monthly by automating data analysis, cleaning, and reporting tools.
- Developed an in-house facial recognition-based attendance system with incredibly low latency and 100% accuracy.
Senior Data Scientist | ML Engineer
SHOHOZ
- Implemented user segmentation with clustering, reducing marketing spend.
- Built fraud detection data pipelines, reducing fraudulent activities by over 50%.
- Architected data solutions for the government-backed Corona Tracer BD.
Machine Learning Engineer
Auleek
- Developed an application to detect architectural components of a floor plan using deep learning.
- Created an application to determine whether a component can be placed in a floor plan.
- Automated training and prediction of floor plan images.
RND Software Engineer
REVE Systems
- Engaged in full-stack development for a government e-vet platform.
- Improved the packet analysis tool for high-throughput networks.
- Developed the application with technologies like Java, Hibernate, JPA ORM, and Thymeleaf.
Data Scientist | ML Engineer
Thread Equation PTE Ltd.
- Developed an application powered by Django and Django REST framework.
- Created an NLP-powered application to identify attacks on applications.
- Integrated the machine learning application into the Django application.
Experience
Resume Optimizer
Live in https://airesumeoptimizer.streamlit.app/
New York Vehicles Crash Data Interactive Visualization
Xin-ORM
https://github.com/proafxin/xin• Execute queries on a database.
• Read a database table as a data frame.
• Write a data frame to a database table (still under development).
• Flatten and normalize a data frame with a nested structure.
• Serialize a data frame as a list of Pydantic models.
• Deserialize a list of Pydantic models as a data frame.
OCR Scorer
https://github.com/proafxin/tdm-trackerCorona Tracer
I was the data lead, building data pipelines, analytics, and visualizations. Due to the nature of the pandemic, one challenge was delivering the project within a very tight timeline. Ultimately, we created automatic real-time data pipelines and provided government officials with analytics, reports, and visualizations.
ShiftSmart
https://shiftsmart.com/Additionally, I used a Flask-based back-end application to facilitate intercommunication between the microservices in the pipeline.
DeepCortex
I worked as a data science team lead. We designed and developed the data science-related back end of DeepCortex.
Profiling Users Through Clustering
Clustering was one of the steps of that pipeline. K-means was used to cluster users, and k-means++ was used as a seeding initialization technique to improve the clustering quality.
Facial Recognition API
One of the challenges was to persist the data involved in this application.
Food Recommendation & Trend Analysis
Before feature engineering and tuning, matrix factorization was used as the primary choice model. I also utilized data mining and machine learning techniques for trend analysis.
Fraud Detection
Fifa Simulation in APIs
https://github.com/proafxin/football_managerIn testing, the code has 100% coverage. Continuous integration is used to make sure no bad code is being pushed or merged.
I used Tox and GitHub Actions to ensure the project can be successfully deployed on various platforms and environments.
Dashboard for Cyberattacks
Pandas Extras
https://github.com/proafxin/pd-extrasData frame-to-database is meant to remove all the extra steps from this writing process. Currently, the goal is to support SQL and NoSQL databases, including data warehouses such as Google BigQuery or Apache Cassandra. For SQL databases, SQLAlchemy is used internally to generalize all SQL database connections.
ETL Job Scheduling with Apache Airflow
https://github.com/proafxin/airflowInternally, SQLAlchemy and PyMySQL are used to connect to the database and communicate with it for reading and writing data. Pandera validates a Pandas data frame created from data retrieved in the SQL query.
Bug Tracker
https://github.com/proafxin/bug-trackerThe app is currently under development to reach 100% test coverage, with the minimum functionalities of creating and editing stories or bugs. The application can also be used as a Docker container. Poetry and Tox are used internally to maintain dependencies.
Seeding Methods in K-means Clustering
https://github.com/proafxin/seeding-kmeansReffer: An Open-source Bibliography Management Solution
Global Entity Disambiguation
https://github.com/proafxin/disambiguationInstead of using massive models to jointly encode signals, I factorized the task into orthogonal channels, each handled by a frozen specialist encoder (BGE-large and RoBERTa-large). Keeping these vectors separate via concatenation enabled an efficient 8.4M-parameter head to jointly evaluate mention geometry. To eliminate bottlenecks, I implemented a window-partitioning design that bounds encoding to linear O(NK) complexity. I then built a cross-window cluster-matching layer using attention pooling to stitch entities across distant windows via a quotient graph, achieving optimal O(C^2) cluster complexity (C << N). My system achieved a CoNLL F1 score of 0.8533 on the CoNLL-2012 benchmark using frozen backbones, drastically reducing token comparisons while matching the performance of fully fine-tuned models.
Certifications
AWS Cloud Technology Consultant
Amazon Web Services
AWS Cloud Solutions Architect
Amazon Web Services
AWS Fundamentals Specialization
Amazon Web Services
IBM Data Science Professional Certificate
IBM | via Coursera
Skills
Libraries/APIs
Pandas, SciPy, NumPy, TensorFlow, PyTorch, OpenCV, SQLAlchemy, Asyncio, Python Asyncio, API Development, Pydantic, Matplotlib, Scikit-learn, REST APIs, PySpark, PyMongo, Pyodbc, PyMySQL, Shapely, OpenAI API, Folium, Back-end APIs, GDAL, GDAL/OGR, Libraries/APIs, React, Claude API, Node.js
Tools
Pytest, Git, Seaborn, Jira, Coverage.py, Apache Airflow, GIS, Plotly, Jupyter, AWS Command Line Interface (CLI), Open Neural Network Exchange (ONNX), BigQuery, ChatGPT, Uvicorn, Grafana, Amazon Simple Queue Service (SQS), Amazon Simple Notification Service (SNS), Amazon CloudWatch, AWS SDK, Amazon Virtual Private Cloud (VPC), GRASS GIS, Terraform, GitHub, AI Prompts, Claude, Claude Code, Amazon Elastic Container Service (ECS)
Languages
C++, Python, Java, SQL, Python 3, JavaScript, TypeScript
Frameworks
Django, Django REST Framework, Alembic, Django Ninja, Flask, Spark, Streamlit, Agentic Frameworks, LightGBM, LangGraph
Paradigms
Scrum, Object-oriented Programming (OOP), Business Intelligence (BI), Socket Programming, REST, Microservices, Test-driven Development (TDD), Testing, Asynchronous Programming, Code Refactoring, Continuous Integration (CI), ETL, RESTful Development, Unit Testing, Continuous Delivery (CD), Real-time Systems, Data-driven Methodology, Distributed Computing, Automation, Scalable Application, Back-end Architecture, Templating, Automated Testing, API Architecture
Platforms
Jupyter Notebook, Amazon Web Services (AWS), AWS Lambda, Google Cloud Platform (GCP), Azure, Apache Kafka, Docker, NVIDIA CUDA, Kubernetes, Ollama, Amazon EC2
Storage
MongoDB, Amazon S3 (AWS S3), NoSQL, MySQL, Databases, Data Pipelines, Redis, Amazon DynamoDB, JSON, PostgreSQL, Microsoft SQL Server, Data Validation, Google Cloud, ScyllaDB, SQLite
Other
Clustering, Code Coverage, Machine Learning, Data Science, Deep Learning, Data Engineering, Recommendation Systems, Google BigQuery, Back-end, QA Automation, Architecture, Software Design, SaaS, API Integration, FastAPI, Poetry, Tox, Containerization, Code Review, Supervised Machine Learning, Async/Await, Generative Artificial Intelligence (GenAI), Large Language Models (LLMs), Web Scraping, Software Architecture, Amazon RDS, Back-end Development, Recurrent Neural Networks (RNNs), Data Visualization, Data Analytics, Data Mining, Agile Sprints, Multithreading, APIs, CI/CD Pipelines, Natural Language Processing (NLP), Document Parsing, Document Processing, ETL Tools, Data Warehousing, Pandera, DataFrames, Generative Pre-trained Transformers (GPT), GeoPandas, RESTful Microservices, Workflow, RESTful Services, Spatial Analysis, Visualization, Integration, Artificial Intelligence (AI), RESTful Web Services, Business Services, Clustering Algorithms, K-means Clustering, NVIDIA TensorRT, Tensorrt, ONNX Runtime, Optical Character Recognition (OCR), EasyOCR, Tesseract, Image Processing, Convolutional Neural Networks (CNNs), Neural Networks, Retrieval-augmented Generation (RAG), Software Engineering, OpenAI, Gunicorn, GitHub Actions, Prometheus, Middleware, OpenAI SDK, OpenAI GPT-4 API, OpenAI GPT-3 API, Open-source LLMs, Meta Llama, Dify, Information Retrieval, Atlas, Asyncpg, Polars, Text-to-Speech (TTS), Speech-to-Text (STT), Image Generation, Moderation, AI Agents, Servers, Dagster, DuckDB, Bokeh, Algorithms, Real-time Data, Real-time Computing, Real-time Vision Systems, Data Analysis, CycleGAN, Generative Adversarial Networks (GANs), Analytical Dashboards, Cloud Computing, Big Data, Collaboration, System Architecture, Security Engineering, AWS Big Data, Solution Architecture, AWS Cloud Security, Cloud Infrastructure, Scalable Architecture, Back-end Performance, Scalable Web Services, Full-stack, Startups, Scalability, Full-stack Development, Data Cleaning, Front-end, Rasterio, Remote Sensing, Geospatial Data, Geospatial Analytics, QGIS, Object Detection, Video Processing, Infrastructure as Code (IaC), Monitoring, Prompt Engineering, PDF, Pdf to text, Libraries, SDKs, Static Libraries, Communication, Mentorship, Troubleshooting & Issue Resolution, Knowledge Graphs, AI Engineering, LangChain, Agentic AI, Agentic AI Systems, Gradient Boosting, Gradient Boosted Trees, Reinforcement Learning, Transformers, RoBERTa, Embedding Models, Machine Learning Operations (MLOps), Supabase, Data Classification, AI Data Classification, Data Modeling, Data Transformation, Technical Architecture, Agentic Coding, Agentic RAG Systems, Agentic Workflow Design, Bun, Claude Design, Recall, Enterprise AI, Transcription, Small Language Models (SLMs), RAG Architecture, Chatbots
How to Work with Toptal
Toptal matches you directly with global industry experts from our network in hours—not weeks or months.
Share your needs
Choose your talent
Start your risk-free talent trial
Top talent is in high demand.
Start hiring