I'm a Data Scientist with 4+ years of experience building scalable, production-grade AI and deep learning systems. My expertise spans across insurance, retail, and media industries, with a focus on optimizing model performance and integrating end-to-end machine learning pipelines.
- M.S. in Computer Science | University of Texas at Arlington (2023 - 2024)
- B.E. in Information Science and Engineering | Visvesvaraya Technological University (2013 - 2018)
- Machine Learning & Deep Learning: Production-grade ML systems, BERT, GRU, LSTM
- Distributed Computing: Parallel Processing, Multi-node Systems, GPU Programming
- Big Data Processing: Apache Spark, Kafka, Redis, MongoDB
- MLOps: Docker, Kubernetes, AWS, GCP, Azure
- Languages & Frameworks: Python, C/C++, PyTorch, TensorFlow, CUDA
Python, PyTorch, Float8 Optimization(In Progress)
- Contributed to open-source development of COAT (Compressing Optimizer States and Activations)
- Implemented memory-efficient FP8 training utilizing dynamic range expansion
- Enhanced float8 data type utilization during training processes
- View Project
C, MPI, OpenMP
- Developed a high-performance parallel solver for 3D heat equations
- Implemented multi-node computation across X, Y, and Z dimensions
- Conducted comprehensive scaling analysis on UTA cluster
- Achieved significant performance improvements through parallel processing
C, Cache Optimization
- Optimized single-core matrix multiplication using advanced algorithms
- Achieved 50%+ of theoretical GFLOPS through cache optimization
- Implemented Goto and Van de Geijn algorithm with six loop orderings
Python, Node.js, gRPC
- Implemented various distributed consensus algorithms:
- Paxos Algorithm Simulation
- Raft Consensus Protocol
- Decentralized & Distributed Locking
- Vector Clocks & Lamport's Logical Clock
- Created cross-language compatible distributed store
- View Project
(Sep 2023 - Jan 2024)
- Engineered LSTM models for smartwatch analytics
- Implemented robust fall detection systems using PyTorch Lightning and MLFlow
(Mar 2022 - Jan 2023)
- Led a team on automating pharmaceutical report analysis via combination of heuristics with the collaboration of subject matter experts and BERT Zero-Shot Classification.
- Reduced the report evaluation process from 1 month to 1 week.
- Implemented dynamic PROMPT/LABEL techniques
(Jan 2019 - Jan 2022)
- Deployed prediction model on News Articles predicting the top 10 articles published on platform for the day with 80% accuracy.
- Reducing load on Redis, Kafka, and MongoDB by developing FAISS vector-similarity API with a scalable design providing sub 8ms response time.
- Leveraged SparkML clustering algorithms for segmented recommendations on News article platform which boosted the average CTR from 1.5 to 1.8.
- Implemented real time preprocessing services using Faust( a python streaming library) which reduced preprocessing cost by 50%.
- Consulted a title insurance organization as deep learning engineer to Train and Deployed a GRU-based NER model on Property legal docs by collaborating with domain experts, achieving 82% accuracy and reduced training time by 30% with PyTorch DDP on multi GPU system.
- Built scalable model-serving pipelines using TorchServe
- Email: [email protected]
- LinkedIn: MirMustafaAli
- Location: Arlington, TX
Currently seeking opportunities in Machine Learning Engineering and Data Science roles