Cisco moved quickly to answer a problem many enterprises face: dense model training demands large, steady streams of data and predictable hardware that fits existing facilities.

Karishma Mandal leads products that align Cisco UCS servers with Nutanix Cloud and the N-AI product category, certifying configurations so customers can deploy validated clusters without lengthy redesign. The result shortens the path from procurement to active training and reduces the risk that heavy compute jobs will outstrip local resources.

Her work centered on removing friction. Engineers receive pre-validated racks that combine GPU acceleration, orchestration platforms, and hardened management tools. One simple action connects a Pod to an existing fabric and initiates configuration through Cisco Intersight. That automation cuts manual configuration work and prevents the kinds of errors that stall training runs. Mandal framed the effort in practical terms: “Every choice ripples across the organization. You must weigh the long tail of each design.” Her prior achievements at major firms gave her a clear sense of which defaults would produce reliable, repeatable outcomes for enterprise operators.

Engineering the secure factory stack

Mandal directed integration across Intersight and Nutanix API technologies to create what her team calls the Secure Factory stack. Automation for data ingestion, search, and movement into compute accelerators addressed the most persistent bottleneck in model work: getting curated datasets to the GPU. Her team automated ingestion and indexing so scientists spent less time hunting for usable samples and more time on experimentation.

Workload protection received equal priority. Cisco embedded telemetry and traffic analysis within each cluster so abnormal behavior shows up early. Hypershield and AI Defense layers observe node interactions and surface anomalies during training. These protections preserve sensitive records while enabling high throughput. Mandal described the stack this way: “Good infrastructure should work quietly and stay out of the way.” Engineers who once spent days staging data now move directly to model cycles while governance and logging continue behind the scenes.

Unified management and measurable results

Mandal championed a bridge between Cisco Intersight and Nutanix Prism that creates a unified cloud operating model. IT teams can view a global fleet of clusters from a single console, schedule deployments, and automate Day 0 through Day 2 tasks. Early customer trials indicated an operational efficiency gain of roughly thirty to forty percent in deployment and management time. Those savings come from fewer manual handoffs, predictable configuration scripts, and one source of truth for cluster state.

Architectural choices included modular expansion of CPU, GPU, and storage resources so capacity grows without reworking the entire cluster. Customers retain local control of data while gaining the throughput required for large training jobs. Mandal oversaw validation procedures that ensure each module behaves under load, mirroring the stress cases large enterprises present. Her record at other firms, where she designed forecasting and automation systems that yielded hundreds of millions in operational value, informed the careful validation cadence used at Cisco.

Pods as repeatable building blocks

Cisco’s Pod concept simplifies heavy training by packaging compute, networking, storage, and software into repeatable units. Each Pod ships with GPU resources, low-latency switches, enterprise storage, and a curated software stack, including orchestration and monitoring platforms. Adding capacity becomes a matter of placing an additional Pod and bringing it under unified management, a method that avoids the long integration cycles enterprises once expected.

Newer Pod versions include next-generation processors and SmartNICs that maintain steady data flow to accelerators. Software for rapid search across large datasets reduces staging time and improves model iteration speed. Mandal’s team tested Pods across varied environments and documented recovery and scaling behavior, creating playbooks that system administrators can follow under operational pressure.

Global deployments and strategic context

Large programs adopted Cisco’s architecture in several regions where national regulations and data residency rules matter. Deployments ranged from greenfield Tier-III facilities to government-backed research campuses, demonstrating that validated hardware and unified software can meet strict local constraints while delivering dense compute. Mandal’s product work fed directly into those deployments through validated designs and hands-on runbooks.

Her product management role emphasized systems thinking over feature lists. She translated engineering requirements into validated products and ensured field teams had the documentation and automation required for repeatable success. Colleagues credit her with making “complexity readable,” reflecting how her teams reduced operational cognitive load and shifted attention from plumbing to modeling.

Why the work matters

Enterprises require predictable, testable systems when they run large training tasks. Mandal’s approach reduced unknowns at purchase, deployment, and scale. Customers gain hardware that fits into existing racks and software that configures nodes automatically, so the time from delivery to production shortens. That predictability produces measurable savings in time and personnel effort and lowers the chance that expensive compute sits idle.

Her leadership combined systems validation, cross-team orchestration, and a focus on user workflows. The result is infrastructure that supports heavy training while letting organizations keep control of sensitive information. Her career arc, moving from design frameworks that produced major operational wins to product leadership at a global networking firm, gives weight to the claim that her work matters at enterprise scale.