Md. Mahir Ashhab

About Me

I’m a Ph.D. student in Computer Science at the University of Virginia. My current research focuses on L4S (Low Latency, Low Loss, Scalable throughput) deployment, under the supervision of Professor Hyojoon Kim.

My broader research interests lie in networking and systems, with previous work involving memory management for LLM inference workloads, edge cloud computing, and social media content moderation with NLP. I’m passionate about solving real-world performance and reliability challenges through clean systems design.

Research Interests

• Low Latency Network Protocols (e.g., L4S)
• Systems for Large Language Model (LLM) inference
• Edge computing & distributed systems
• Serverless computing and resource management

Projects

• Deploying L4S in Testbed Environments — Working on configuring mixed AQM environments for realistic L4S evaluation using DualPI2 and legacy FIFO nodes.
• Lookahead Caching for LLM Inference — Designed a caching policy optimizing memory reuse across structured prompt workloads.
• Rule Violation Detection in Social Media — Used NLP to identify policy-breaking posts with contextual reasoning and explanations.
• Lookahead Caching Policy for KV Cache Management (2024)
  Tool: SgLang, Python
  A comprehensive study evaluating the effectiveness of lookahead caching versus LRU in improving KV cache hit rates during LLM inference.
• Neural Abstractive Text Summarization with Sequence-to-Sequence Models (2024)
  Tool: BiLSTM, Python, TensorFlow
  Explored sequence-to-sequence architectures for generating abstractive summaries of long-form textual content.
• Human-in-the-Loop Annotation for GitHub Comment Classification (2023)
  Tool: BERT, Python, PyTorch
  Investigated the impact of human-in-the-loop feedback for improving classification accuracy of LLMs on GitHub issue comments.
• Sentiment Analysis on Bangla Text (2023)
  Tool: BERT, Python, PyTorch
  Conducted a large-scale study on using transformer-based models to analyze sentiment in Bangla social media text.
• Laptop Components Classification for E-Waste Management (2023)
  Tool: Python, CNN models, PyTorch
  Created a comprehensive dataset and CNN pipeline for classifying laptop components to facilitate automated e-waste recycling.

Publications

• An Extensive Photographic Dataset to Classify Laptop Components for Automating E-Waste Management
  Data in Brief, Elsevier, Volume 57, 2023 – Accepted and Published
• A Comparative Analysis of Deep Learning Approaches in Bangla Document Categorization
  ICCIT 2023 – 26th International Conference on Computer and Information Technology, ISBN: 979-8-3503-5901-5 – Accepted and Published
• Detecting Pneumonia from X-Ray Images Using Deep Convolutional Neural Networks
  IBDAP 2023 – 4th International Conference on Big Data Analytics and Practices, ISBN: 979-8-3503-0019-2 – Accepted and Published
• An Empirical Study to Analyze the Impact of Instagram on Students’ Academic Results
  TENSYMP 2020 – IEEE Region 10 Symposium, ISSN: 2642-6102 – Accepted and Published
• Execution Delay-aware Task Assignment in Mobile Edge Cloud and Internet Cloud
  STI 2019 – International Conference on Sustainable Technologies for Industry 4.0, ISBN: 978-1-7281-6099-3 – Accepted and Published

📚 View all publications on Google Scholar

CV & Contact

📄 View my CV
📧 Email: ftm2nu [at] virginia.edu
🔗 GitHub | LinkedIn