MAHESH PABBA
Customer Delivery Architect · AI & Cloud Infrastructure
Professional Summary
Seasoned IT Infrastructure Leader with 22+ years of experience in data center computing, enterprise architecture, and technical solutions delivery. Specialized in designing high-performance AI training and inference infrastructure, and building full-stack automation platforms for enterprise customers.
Expert in architecting scalable AI factories on Cisco UCS platforms with NVIDIA GPUs, OpenShift/Kubernetes clusters, and high-speed RDMA networking. Proven track record in infrastructure automation — from Ansible/Terraform provisioning pipelines and Cisco NSO-driven network device upgrades, to GenAI platforms (LangChain, LangGraph, RAG) and fine-tuned LLM deployments. Built production systems across the full stack: FastAPI backends, Angular frontends, Kafka message buses, and containerized DevOps stacks.
Professional Experience
Customer Delivery Architect — Data Center & AI Infrastructure
Cisco
- ▸Designed and deployed high-performance AI training & inference clusters with NVIDIA GPUs, OpenShift, and RDMA networking (PFC, ECN, RoCE v2) on Cisco UCS X-Series platforms.
- ▸Led end-to-end AI infrastructure implementations using NVIDIA GPU Operator, Network Operator, and Node Feature Discovery (NFD) managed through Cisco Intersight.
- ▸Developed Python + LangChain agents; fine-tuned Llama2 on Intersight operational data for domain-specific inference; built autonomous AI-driven Day-2 operations workflows.
- ▸Designed and built DCAF (DataCenter AI Frameworks) — open-source GenAI platform: FastAPI + LangChain/LangGraph backend, React 18 frontend, Ollama local LLM inference, MongoDB Atlas Vector Store for multi-modal RAG (PDF/DOCX/PPTX/images/YouTube), NATS JetStream, OpenVINO INT4/INT8 for GPU-free environments, Vosk offline speech-to-text, OIDC auth, and NVIDIA NIM integration for Cisco UCS X AI Pod.
- ▸Published open-source AI Lab training notebooks — ML fundamentals, synthetic instruction dataset generation (JSONL for LLM fine-tuning), LoRA fine-tuning on Intersight domain data, GPT-2 full fine-tune, and transformer model from scratch.
- ▸Co-authored Cisco Validated Design (CVD): FlashStack for Enterprise RAG Pipeline with NVIDIA NIM, NIM Operator, and RAG Blueprint — validated enterprise RAG architecture on Cisco UCS X-Series with Pure Storage and NVAIE. (Published May 2025)
- ▸Built and published ansible-openshift — end-to-end Ansible automation for bare-metal OpenShift deployments via Red Hat Assisted Installer API: Redfish BMC for ISO mounting, NMState static network config generation, OLM operator installation (NFD, GPU, LVM), state-aware resumable execution, and Day-2 node add/remove operations.
- ▸Automated compute provisioning for Cisco UCS, Fabric Interconnects, and OpenShift clusters using Terraform, Ansible, and Python — significantly reducing customer deployment time.
- ▸Deployed and commissioned Azure Stack Hub Appliance on Cisco UCS hardware; configured BGP on Cisco routers for on-premises to Azure Express Route connectivity; enabling hybrid cloud SAP and enterprise workloads.
- ▸Contributed to Cisco BPA (Business Process Automation) platform — an internal automation product built with an Angular web frontend, multiple Python/Java backend microservice controllers, Cisco NSO for Nexus and IOS-XE device upgrades, Kafka message bus, and a containerized DevOps stack (Docker, Kubernetes, MongoDB, Kong API Gateway, Grafana).
- ▸Developed a Python FastAPI Webex bot for self-service network automation — accepts natural language commands via Webex messaging, routes requests through a Python backend to execute against network devices and Cisco infrastructure, enabling teams to perform Day-2 operations without CLI access.
- ▸Built and maintained Kubernetes/OpenShift clusters across AWS, Azure, and on-prem with Kubespray; supported customer AI training and inference workloads at scale.
- ▸End-to-end SAP HANA TDI design, hardware sizing, HA and DR implementations on Cisco UCS for enterprise customers.
- ▸Technical lead for hybrid cloud/on-premise SAP landscape strategies; deployed SAP HA/DR on Azure using ARM templates, Express Routes, and BGP.
SAP Architect
Numbers Only · Client: Cisco, Raleigh NC
- ▸Installed SAP HANA on Cisco UCS Rack and Blade Servers with EMC, NetApp, and HP 3PAR storage configurations; sizing BW on HANA and migration planning.
- ▸SAP OS/DB migration (HP-UX to RHEL Linux) via Oracle-to-Oracle Online Migration with near-zero downtime on multi-terabyte systems (ECC, BW, SCM, HR).
- ▸VMware HA for ACS (enqueue/message server high availability); re-implemented SolMan 7.1.
Sr. Consultant
Deloitte
- ▸Installed parallel GRC 10.0 landscape on NetWeaver 7.3; migrated from GRC 5.3 to 10.0 and enabled AC, PC, RM modules.
- ▸Established BOBJ to SAP HANA database connectivity for the BI development team.
Senior Consultant
CapGemini
- ▸Managed multiple concurrent SAP projects (R/3 ECC 6.0, CRM 5.0, BI 7.0, EP 7.0, BPC, BO XI 3.1) for US and India clients simultaneously.
- ▸OS/DB migrations, Unicode conversions, upgrades, system copies, and SAP-to-SAP & SAP-to-3rd-party integration activities.
- ▸MSSQL database administration, workload analysis, performance tuning, memory profile parameter adjustments.
Earlier Experience (2003 – 2008)
| Period | Role | Company / Client |
|---|---|---|
| 2007–2008 | IT Consultant | Reliance Retail Limited, Mumbai — SAP Portal, BI, R/3 |
| 2005–2007 | Associate BASIS Consultant | Wega Solutions, Bengaluru — SAP BASIS administration |
| 2003–2005 | System Administrator | Mother Dairy India Ltd, Hyderabad — NT/2000/2003, 100+ users |
Technical Skills
| Cloud | AWS · Azure · GCP · Azure Stack Hub |
|---|---|
| Orchestration | Kubernetes · OpenShift · Docker · Docker Swarm · Helm |
| Automation | Python · Ansible · Terraform · Bash · Redfish/BMC |
| AI / GenAI | LangChain · LangGraph · LLM Fine-tuning · LoRA · NVIDIA GPU Op · RDMA/RoCE · Ollama · OpenVINO |
| Backend / APIs | FastAPI · Python · Kafka · NATS · NSO · Kong API Gateway |
| Frontend | React · Angular · Vite · TailwindCSS · DaisyUI |
| Networking | BGP · Cisco NSO · PFC · ECN · RoCE v2 · Cisco Fabric Interconnects · Webex |
| Databases | MongoDB · SAP HANA · Oracle · MSSQL · MySQL · Redis |
| Storage | EMC · NetApp · Pure Storage · VAST · HP 3PAR |
| Servers | Cisco UCS (Rack & Blade, X-Series) · HP · Dell |
| OS | RHEL · SUSE Linux · HP-UX · Windows Server |
| SAP | HANA · S/4HANA · ECC · BW · GRC · PI/PO |
Education & Certifications
Publications, CVDs & Projects
Cisco Validated Design (CVD) · Co-Author · Published May 2025
Validated architecture for enterprise-grade Retrieval Augmented Generation pipelines on Cisco UCS X-Series with Pure Storage FlashArray/FlashBlade, NVIDIA AI Enterprise (NVAIE), NIM microservices, NIM Operator, and Red Hat OpenShift — jointly engineered by Cisco, Pure Storage, and NVIDIA.
Open-Source GitHub Project · Jupyter Notebooks · Active
Hands-on AI/ML notebooks organized in three tracks: ML Basics (supervised/unsupervised learning, transformer from scratch), Datasets (synthetic instruction dataset generation, Intersight API dataset, PDF extraction, web scraping for LLM fine-tuning), and Fine Tuning (LoRA fine-tuning on domain-specific data, GPT-2 full fine-tune, unsupervised fine-tuning, end-to-end model creation).
Open-Source GitHub Project · Full-Stack GenAI Platform · Containerized
Production-grade GenAI platform for datacenter use cases. Stack: React 18 + Vite frontend, FastAPI + LangChain + LangGraph backend, MongoDB Atlas vector store, NATS JetStream message queue. Inference backends: Ollama (local LLMs), OpenVINO INT4/INT8 (CPU-only), llama.cpp/GGUF over NATS, and NVIDIA NIM on Cisco UCS X AI Pod (remote mode). Features include multi-modal RAG (PDF, DOCX, PPTX, images, YouTube), semantic caching, OIDC auth (Cisco SSO + Duo Security), LangGraph ReAct agent orchestration, visual workflow builder, and custom Python tool authoring.
Open-Source GitHub Project · Ansible · Red Hat OpenShift
End-to-end Ansible automation for deploying OpenShift clusters on bare-metal via the Red Hat Assisted Installer API. Features: Redfish BMC automation for discovery ISO mounting, NMState static network configuration generation per server, OLM operator management (NFD, GPU, LVM, NMState), state-aware resumable deployments (safe to interrupt and re-run), and Day-2 operations for adding/removing worker nodes.
Use your browser's print dialog → Save as PDF for best results.