Mufakir Qamar Ansari
Data Scientist · AI Researcher & Systems Architect
Data Scientist with 9+ years of combined research and industry experience in predictive modeling, statistical analysis, and applied machine learning. Builds end-to-end ML and ETL pipelines and applies generative AI, NLP, and deep learning to complex, high-dimensional datasets on distributed computing platforms. Translates analytical findings into actionable insights for cross-functional stakeholders and executive leadership across diverse domains.
Skills & Technical Tools
Python, SQL, PySpark, R, C++, CUDA
Predictive Modeling, Generative AI, Deep Learning, NLP, Time Series Forecasting, Computer Vision, A/B Testing, Causal Inference, Statistical Modeling
PyTorch, TensorFlow, Scikit-learn, LangChain, Hugging Face, Transformers
AWS (SageMaker, S3), Azure Databricks, Apache Airflow, Docker, MLflow, Git, SLURM HPC
PostgreSQL, MySQL, MongoDB, ChromaDB, Neo4j, Tableau, Optuna, ArcGIS Pro
Formal Academic Degrees
Master of Science in Computer Science & Engineering
Aug 2023 – Aug 2025Coursework and research: Advanced Deep Learning, High Performance Distributed Computing, Statistical Machine Learning, and Parallel Algorithms.
Bachelor of Technology in Electronics & Communication Engineering
Jul 2009 – Jul 2013Premier Institute of National Importance. Core training in digital signal processing, communication theory, stochastic modeling, embedded systems, and discrete mathematics.
Research Experience
Research Assistant · Wright State University
Jan 2026 – Present- Architected a 19-stage HPC ETL pipeline that processed 356 RNA-Seq samples and 15M+ read pairs by orchestrating SLURM-based parallel execution on the OSC supercomputer, with full reproducibility via AWS S3 and Azure Databricks.
- Engineered a generative AI benchmarking pipeline for HDLSS synthetic tabular data by training 7 deep generative models (CTGAN, TabDDPM, FlowMatching, TVAE) class-conditionally and validating output fidelity through Wasserstein distance, KS likeness, and Train-Synthetic-Test-Real (TSTR) classification across 7 classifiers.
- Developed DDFF, a distributional feature selection framework that achieved 99.7% dimensionality reduction on high-dimensional datasets by quantifying class-conditional PMF divergence in O(pn) time; results accepted at IEEE ICMLA 2026.
Research Assistant · The University of Toledo
Aug 2023 – Apr 2025- Delivered predictive crash-severity models for the Ohio Department of Transportation by processing 37,000+ geospatial records and engineering full-stack ensemble classifiers (Random Forest, XGBoost, LightGBM), producing a published state report (FHWA/OH-2025/19).
- Reduced deep learning training time by 50% (2x speedup) and accelerated GPU execution by 8.5% by optimizing distributed HPC workloads across SLURM-managed GPU clusters, while building a predictive scheduling model that cut the cluster's carbon footprint by 5%.
Professional Industry Experience
Technical Lead · Orcinus IT Solutions Private Limited
May 2018 – Dec 2022- Reduced data pipeline latency by up to 35% for 7 SaaS enterprise clients by designing and deploying end-to-end ML solutions and scalable ETL workflows using Apache Airflow, PySpark, and Azure.
- Improved model reliability by 30% across production deployments by building a comprehensive MLOps framework enforcing version control, automated monitoring, and governance.
Senior Engineer, then Consultant · MyFajir IT Solutions Private Limited
Sep 2017 – Mar 2020- Processed 800+ daily transactions at 99.9% uptime by architecting production-grade RESTful APIs (Flask/FastAPI) and interactive analytics dashboards for cross-functional stakeholder decision support.
- Reduced client operational overhead by 20% by engineering a cloud-based ERP platform on Azure with SQL-driven relational database integration and agile sprint delivery via JIRA.
Business Operations Lead · PanunKart.com
Aug 2013 – Sep 2017- Drove 150% sales growth by designing and implementing a data strategy grounded in A/B testing and market segmentation analysis, directly improving data-driven decision-making across the organization.
Featured Engineering Projects
High-Performance Computing Pipeline for Ebola RNA-Seq Analysis
GitHub open_in_newArchitected an end-to-end, 19-stage ETL pipeline on the OSC supercomputer via SLURM array parallelization, processing 356 samples and over 15 million read pairs from the 2014 West African Ebola outbreak. Engineered robust dual-quantification workflows (HISAT2, Kallisto) and orchestration logic, ensuring seamless terabyte-scale data processing and generating cross-validated transcriptomic analytics.
Privacy-Preserving AI Agent & Knowledge Graph Engine
GitHub open_in_newEngineered a highly secure, cross-platform AI agent ecosystem (Tauri/Rust, FastAPI, Next.js) featuring robust AES-256-GCM encryption and local LLM inference, guaranteeing strict data sovereignty. Architected a real-time Neo4j Knowledge Graph powered by custom NLP pipelines to continuously ingest, map, and analyze complex cross-platform user activity into personalized insights.
Production LLM & RAG Implementation: Hybrid Optimization System
GitHub open_in_newOrchestrated an end-to-end LLM system featuring a RAG pipeline and ChromaDB vector database for complex financial data processing and entity extraction. Evaluated and benchmarked multiple foundation models (Phi2, Mistral 7B, Zephyr 7B) by optimizing prompt engineering, vector search latency, and production monitoring protocols.
Research Papers & Academic Service
- M. Ansari, L. Popp, T. Ghosh, "Distributional Discriminative Feature Filtering: A Scalable Framework for High-Dimensional Low-Sample-Size Biomedical Data", Accepted in IEEE ICMLA 2026.
- M. Q. Ansari and M. Ansari, "Disentangling Architecture, Strategy, and Adaptation in Day-Ahead Electricity Price Forecasting: A Multi-Model Benchmark on Volatile US Markets", Under Review, Computational Economics (Springer Nature).
- M. Q. Ansari and M. Ansari, "DynaCut: Bounding Classical Memory in Quantum Circuit Knitting via Resource-Aware Tensor Network Reconstruction", Under Review, Quantum Machine Intelligence.
- M. Q. Ansari, "Dy-Part: A Dynamic, Noise-Aware Scheduler for Optimizing Hybrid Quantum-Classical Algorithms", Under Review, Quantum Machine Intelligence. Preprint available: DOI: 10.21203/rs.3.rs-8041248/v1.
- M. Q. Ansari, et al., "From Text to Returns: Using Large Language Models for Mutual Fund Portfolio Optimization and Risk-Adjusted Allocation", Preprint available: DOI: arXiv:2512.05907v1.
- M. Q. Ansari, et al., "High-Sensitivity Detection of Invasive Ductal Carcinoma via Domain-Specific SimCLR Pre-Training", Under Review, Signal, Image and Video Processing. Preprint available: DOI: 10.21203/rs.3.rs-8031909/v1.
- E. Chou, R. Rai, M. Ansari, S. Khanal, A. Davenport, "From Data to Action: Leveraging Machine Learning/Artificial Intelligence to Guide Proactive Pedestrian and Cyclist Safety Initiatives", Ohio Department of Transportation Final Report FHWA/OH-2025/19, August 2025. Available: SJN 136837.
- Hasan, T., Hossain, A., Ansari, M. Q., & Syed, T. H. (2025), "Enhanced Intrusion Detection in IIoT Networks: A Lightweight Approach with Autoencoder-Based Feature Learning", Proceedings of the 10th International Conference on Internet of Things, Big Data and Security (IoTBDS 2025), pp. 207-214.
- M. Q. Ansari, et al., "Racing to Idle: Energy Efficiency of Matrix Multiplication on Heterogeneous CPU and GPU Architectures", American Journal of Computer Science and Technology (AJCST), ISSN Online: 2640-012X; ISSN Print: 2640-0111, 2025. Preprint: DOI: 10.21203/rs.3.rs-7890483/v1.
- M. Q. Ansari, et al., "Accelerating Matrix Multiplication: A Performance Comparison Between Multi-Core CPU and GPU", Preprint available: DOI: 10.48550/arXiv.2507.19723.
Peer Reviewer: Engineering Applications of Artificial Intelligence (Elsevier, 43 papers verified), Journal of Pathology Informatics (Elsevier / API), IEEE ICMLA 2026, IEEE ICECCME 2026, and IEEE ACDSA 2027.
Conference Presentations & Symposium Talks
Enhanced Intrusion Detection in IIoT Networks: A Lightweight Approach with Autoencoder-Based Feature Learning
Apr 2025Speaker & Presenter: Mufakir Qamar Ansari. Authors: Tasnimul Hasan, Abrar Hossain, Mufakir Qamar Ansari, Talha Hussain Syed.
Delivered the technical oral presentation for Paper #19 on a deterministic, class-weighted bottleneck autoencoder ($24 \to 6 \to 24$) with edge deployment on NVIDIA Jetson Nano hardware, demonstrating 99.94% Macro F1 with 0.185 ms inference latency for mission-critical Industrial IoT edge networks.
From Data to Action: Leveraging Machine Learning/Artificial Intelligence to Guide Proactive Pedestrian and Cyclist Safety Initiatives
Oct 2024Authors: Rabinson Rai, Mufakir Qamar Ansari, Abiskar Nepal, Eddie Chou.
Presented state-level machine learning frameworks and predictive spatial safety models developed for the Ohio Department of Transportation (ODOT) to evaluate high-risk road corridors, vulnerable road user (VRU) crash severities, and data-driven infrastructure countermeasure prioritization.
Installation, Testing and Data Analysis for a Weather Station
2023Authors: Dr. Stefan Andrei, Harini Doppalapudi, Ramanpreet Singh, Venkata Raghu Vishnumolakala, and Mufakir Qamar Ansari.
Presented sensor deployment architectures, data telemetry validation, and statistical environmental analysis for meteorological monitoring systems supporting community resilience and extreme weather response.