Syed Muhammad Aqdas Rizvi
(smarizvi110)

Lahore University of Management Sciences (LUMS) | Lahore, Pakistan

Relevant Coursework: Networks and Systems, Data Structures/Algorithms, Software Engineering, Operating Systems, Databases, AI & ML, Natural Language Processing.

Karachi, Pakistan

Conducting independent, cross-layer computer science research involving Distributed Systems, Computer Networks, HCI, and AI. Focused on empirical benchmarking, low-level protocol optimization, and publishing findings.

MLSys & Cryptoeconomics (Sentinel-Bench Project)

• Authored a working paper bridging AI Alignment with distributed systems physics, evaluating 9B-parameter LLMs as edge-native governance firewalls. [View Repository]
• Engineered a robust, 840-inference pipeline utilizing Python, DSPy, and Ollama to conduct a strict intra-model ablation on the Qwen-3.5-9B architecture.
• Corroborated a severe “Compute-Accuracy Inversion,” showing that System 2 (Chain-of-Thought) reasoning suffers a 26.7% cognitive collapse rate on complex legal texts, degrading trial-to-trial consensus stability to 72.6%.
• Translated LLM behavioral metrics into hard cryptoeconomic vulnerabilities, proving that the 17x latency overhead of reasoning models critically threatens Byzantine Fault Tolerance (BFT) and exposes networks to Governance Extractable Value (GEV).

Transport-Layer Protocol Modification (CATS Project)

• Conceived, designed, and evaluated CATS, a novel framework endowing TCP with semantic priority awareness, culminating in a sole-author publication on IEEE Xplore. [View Repository]
• Engineered a custom C++ TCP socket implementation within the ns-3 network simulator, utilizing an "Interceptor and Feeder" architecture seamlessly operating alongside the TCP BBR congestion control algorithm.
• Designed a sophisticated two-threshold hysteresis intra-flow fairness mechanism to prevent starvation of low-priority data.
• Executed a rigorous, representative worst-case webpage download simulation, extracting verifiable PCAP traces using a custom TCP Option.
• Proved a 78.7% improvement in First Contentful Paint (FCP) by intelligently reordering packet transmission, radically improving user-perceived QoE with negligible overhead to total load time.

Center for Speech and Language Technologies (CSaLT) | Lahore, Pakistan

• Conducted a comprehensive literature review of Keyword Spotting (KWS) technologies, focusing specifically on Urdu. [View Report]
• Analyzed current acoustic modeling challenges and proposed architectural solutions to advance accessibility in low-resource AI environments, publishing the findings as an independent arXiv preprint.

PhysLab | Lahore, Pakistan

• Investigated Faraday's Law of Electromagnetic Induction; co-authored the official experiment manual and developed MATLAB scripts for automated data processing, both published on the laboratory's platform. [View Project]
• Refined experimental setups for Surface Plasmon Resonance to significantly improve measurement accuracy.

Risk Associates | Karachi, Pakistan

• Translating abstract security and compliance requirements (ISO/IEC 27001 & 42001) into concrete, system-level controls spanning software architecture, data governance, and access control networks.
• Conducting threat modeling and risk assessments for enterprise infrastructures, gaining deep practical insight into real-world distributed systems vulnerabilities and mitigation strategies.
• Assisting in the design of AI Management Systems, providing hands-on exposure to emerging challenges in AI governance, model lifecycle oversight, and secure pipeline architecture.

AI in Healthcare Initiative (AIHI), LUMS | Lahore, Pakistan

• Developed MigraineAid, an applied machine learning platform to help sufferers predict attack occurrences.
• Implemented a high-performance backend in Rust to ensure memory safety and fast data processing for real-time analytics, integrating Firebase and Firestore for scalable data management.
• Contributed to ML model development pipelines and built a secure frontend dashboard in JavaScript for researchers.

TeReSol (for Bank AL Habib Limited) | Karachi, Pakistan

• Engineered reusable micro-frontend components for core banking software utilizing a finite state machine (FSM) to manage strict user workflows and API integrations under high-security banking constraints.