Rahul Verma

Research Interests

Multimodal AI  ·  Distributed & Federated Machine Learning  ·  Agentic AI Systems  ·  AI/ML for Networks & Infrastructure  ·  Foundation Models & LLMs  ·  GPU-Accelerated Computing

Education

• Indian Institute of Technology Madras Sep 2020 – Jun 2023
  MS (by Research), Computer Science & Engineering — CGPA 8.76
  Thesis: Federated Learning for VNF auto-scaling and placement in 5G-and-beyond networks.
  Advisor: Prof. Krishna Moorthy Sivalingam.
• Terna Engineering College, Mumbai University Jun 2014 – May 2018
  B.E., Information Technology — CGPA 8.67

Industry Experience

• Cisco Systems August 2024 – Present
  Software Engineer, Grade 08
  • Worked on Linux kernel–level systems and eBPF-based monitoring for fast vulnerability mitigation and system observability.
  • Implemented IEEE 802.1Qbu TSN Frame Preemption on Cisco Industrial Ethernet 9000 — including firmware changes, QoS integration, and per-port CLI configuration enabling high-priority traffic to preempt non-critical frames.
  • Developed PoE features, firmware boot-time upgrades, and enforced SELinux security policies.
  • Built an LLM-driven automated code review platform with data-centric feedback loops to continuously improve code quality at scale.
  • Designed a Graph RAG agent for automated PHY/port root-cause analysis, combining hardware dependency knowledge graphs with LLM reasoning to reduce engineer triage time from ~60 to ~5 minutes.
• OPLUS India R&D (OPPO / OnePlus) July 2023 – August 2024
  Engineer L3
  • Developed a Magic Eraser–style generative AI feature for intelligent object removal and background synthesis on-device.
  • Built ML-based failure detection pipelines and LLM-assisted log analysis for accelerated root-cause identification.
• Tata Consultancy Services July 2018 – September 2020
  System Analyst
  • Developed enterprise distributed applications with ML-ready data interfaces for predictive analytics and scalable automation workflows.

Research Experience

• VMware University Research Fund, USA  ·  IIT Madras June 2021 – May 2023
  Research Associate  |  Advisor: Prof. Krishna Moorthy Sivalingam
  • Designed a federated learning framework for VNF auto-scaling in multi-domain 5G networks using LSTM/GRU forecasting and Time-GAN synthetic data generation, preserving privacy across network domains.
  • Developed a federated deep learning system for dynamic VNF placement in multi-cloud environments; validated on synthetic and real-world datasets with measurable reduction in over/under-allocation.
  • Program: "Orchestration for network slicing in 5G networks using SDN/NFV."
• Publications
  • Design and Analysis of VNF Scaling Mechanisms for 5G-and-Beyond Networks using Federated Learning

    Rahul Verma, Krishna M. Sivalingam, and Omkar Chavan
    IEEE Access, vol. 12, pp. 129826–129843, Sep. 2024

    paper
  • VNF Placement based on Resource Usage Prediction using Federated Deep Learning Techniques

    Rahul Verma, Krishna M. Sivalingam, and Omkar Chavan
    IEEE Future Networks World Forum (FNWF), Baltimore, USA, 2023

  • Federated Learning approach for Auto-scaling of VNF Resource Allocation in 5G-and-Beyond Networks

    Rahul Verma and Krishna M. Sivalingam
    IEEE 11th International Conference on Cloud Networking (CloudNet), pp. 242–246, 2022

• Teaching Assistant  ·  IIT Madras Apr – Jun 2021
  CS1111: Problem Solving Using Computers
  Conducted labs, held office hours, and graded assignments for an undergraduate programming course.

Projects — Cisco Systems

• IEEE 802.1Qbu TSN Frame Preemption — Cisco Industrial Ethernet 9000
  Cisco Systems · 2024 – Present
  Time-Sensitive Networking requires high-priority frames to preempt in-flight non-critical traffic with microsecond-level latency guarantees — a capability that demands tight co-design across firmware, OS, and network stack. Implemented end-to-end TSN preemption support: platform firmware changes, QoS integration for handling express and preempted traffic queues, per-port CLI configuration, and an auto-enable mechanism that activates preemption only when both link endpoints negotiate support. Validated on Cisco IE9000 industrial switch platforms.
• AI-Powered Automated Code Review Platform
  Cisco Systems · 2024 – Present
  At scale, manual code review misses subtle bugs, security issues, and cross-file regressions — especially when a change in one module silently breaks a dependency elsewhere. Built a system where specialized LLM agents independently analyze each PR for bugs, security, performance, and maintainability, grounded by retrieval over code knowledge graphs and semantic embeddings to reason across file boundaries. Agent findings are aggregated with static analysis signals and ranked by confidence; a developer feedback loop retrains the ranking model over time.
• Graph RAG Agent for Network PHY/Port Debugging
  Cisco Systems · 2024 – Present
  PHY/port issues — link flaps, CRC bursts, SFP failures — rarely have a single-source root cause; diagnosis requires correlating CLI outputs, driver logs, transceiver metrics, and hardware errata across the full Port → PHY → SerDes → ASIC stack. Built a Graph RAG system that models these hardware dependencies as a knowledge graph, enabling an LLM agent to traverse causal paths and pinpoint likely failure points rather than returning semantically similar documents. Reduced average triage time from ~60 to ~5 minutes for complex PHY faults.

Projects — University & Academic

• Extractive Question Answering with DistilBERT (Deep Learning, CPSC-8430)
  Fine-tuned DistilBERT on SpokenSQuAD for extractive QA. Applied Linear LR Decay and Automatic Mixed Precision (AMP) as performance enhancements, improving test F1 from 52.25 → 54.05.
• GAN Comparison: DCGAN vs. WGAN vs. ACGAN on CIFAR-10 (Deep Learning, CPSC-8430)
  Implemented and benchmarked three GAN architectures for image generation evaluated by FID score. DCGAN achieved best FID of 206.19 (vs. WGAN 259.59, ACGAN 281.99), outperforming all variants across 40 epochs.
• GPU TLB Optimization: DUCATI Architecture (GPU Programming, IIT Madras)
  Implemented a simplified DUCATI architecture to extend TLB reach in GPU-accelerated systems using Macsim. Demonstrated fewer TLB evictions than the baseline on GPU workloads (benchmarked via vector_add); CPU-side validated using merge sort.
• Router Architectures and Algorithms (CS6040, IIT Madras)
  Python circuit switching and packet processing simulators; IPv4 classification using Set Pruning Trie.
• Pattern Recognition and Machine Learning (CS5691, IIT Madras)
  Implemented regression, k-NN, Naive Bayes, SVM, and MLFFNN from scratch for image classification.
• Computer Architecture (CS6600, IIT Madras)
  MESI cache coherence protocol simulation; SIMD intrinsics for accelerated image processing; Cachegrind-based matrix multiplication analysis.

Achievements & Recognition

• All India Rank 502 in GATE Computer Science 2020 (97,481 candidates).
• Selected for Amazon ML Summer School 2022.
• Recipient, Late Dr. Smt. Chandrakaladevi Padmasinha Patil Scholarship for outstanding academic performance.
• Recipient, Tata Trust Scholarship for excellent academic performance.
• Completed Deep Learning Specialization — Andrew Ng, Coursera (2021).

Technical Skills