P S Sree Harsha / AI Software Engineering

Engineering Atlas

The systems, technologies and decisions behind the work.

System comparison

Research

Enterprise Internal

Both place controls around model-assisted reasoning: OpsNexus proposes permission-aware graph and vector retrieval, while Hexus documents guarded RCA and ingestion-time redaction.

Problem

OpsNexus

Every function runs its own tooling, dashboards and AI, so teams spend an estimated 60–70% of analytical effort assembling context before deciding anything. A major GPU procurement can move through six functions with every dashboard green and still land hardware before capacity is ready. The failure is not execution; it is that nothing reasons across the domain boundaries.

Project source

Hexus

Incident triage across multi-tenant log streams was manual and slow, with root-cause analysis bottlenecked on tribal knowledge and no safe way to handle sensitive data.

Project source
Project status

OpsNexus

Research

Project source

Hexus

Enterprise Internal

Project source
Architecture

OpsNexus

Existing enterprise systems feed a knowledge and memory graph that agents reason over, producing impact predictions delivered into Teams, and every outcome is written back.

Project source

Hexus

Logs stream in, ML flags anomalies, and an LLM agent reasons out the root cause, all behind RLS, SSO and PII anonymization.

Project source
AI pattern

OpsNexus

LangGraph, Azure OpenAI, Knowledge Graphs, RAG

Project source

Hexus

LangGraph, Azure OpenAI

Project source
Orchestration

OpsNexus

Coordinate the planner, registered MCP tools and durable reasoning steps, with human approval as an explicit gate.

Project source

Hexus

Run RCA as a guarded LangGraph workflow with typed tools and validation around model calls. Provide language-model reasoning within the RCA workflow; generated incident statements need grounding and validation.

Project source
Data

OpsNexus

Connect assets, services and dependencies with a memory of prior decisions, fixes and outcomes.

Project source

Hexus

Store tenant data behind PostgreSQL Row-Level Security. Correct policies and privileged-access controls remain necessary.

Project source
Infrastructure

OpsNexus

Azure OpenAI, Azure AI Search

Project source

Hexus

Azure OpenAI, Kubernetes, Docker

Project source
Human involvement

OpsNexus

Coordinate the planner, registered MCP tools and durable reasoning steps, with human approval as an explicit gate.

Project source

Hexus

Not specified in this project record.

My contribution

OpsNexus

System design, the knowledge and memory graph architecture, agent orchestration, and the Phase-1 pilot plan.

Project source

Hexus

The multi-tenant API, anomaly-detection lifecycle, agentic RCA pipeline, and security and deployment integration.

Project source
Engineering trade-offs

OpsNexus

Engineering trade-offs and rationale

Consumed the existing systems of record instead of building a new data platform. Reading GreenLake, OpsRamp, ServiceNow and SAP through their published APIs avoids replacing the systems of record. Connector development, entitlement mapping and operational support still require engineering work; the design reuses existing platforms rather than introducing a competing data platform. Paired vector retrieval with graph traversal rather than shipping a plain RAG assistant. Vector search finds semantically similar things; graph traversal finds causally connected ones. Vector-only cannot answer what a change affects three hops away, and graph-only cannot read runbooks, tickets and email. Cross-domain operational reasoning needs both, so the hybrid is the technical core of the design. Captured entitlement metadata at ingestion instead of filtering only at query time. If access control is applied late, retrieval can assemble an answer from records the requester is not entitled to see. Binding entitlement metadata at ingestion supports permission-aware retrieval, but enforcement still depends on correct query-time checks, current permissions and tested policies. Delivered recommendations into Teams instead of building another dashboard. Adoption is the real risk on a layer like this, and a portal nobody opens produces no measurable benefit no matter how good the reasoning is. Arriving where people already work is what makes the pilot measurable at all.

Project source

Hexus

Engineering trade-offs and rationale

Enforced tenant isolation in PostgreSQL with Row-Level Security, not in the API layer. Database policies reduce reliance on every API remembering a tenant filter, at the cost of query and operational flexibility. Row-Level Security still needs correct role configuration, policy tests and controls for privileged access; it does not replace a full threat model or security testing. Led with unsupervised anomaly detection (Isolation Forest / LOF) before supervised scoring. Labelled incident data is scarce and biased early on. Unsupervised detection delivers day-one value with no training set; supervised scoring folds in later as labels accumulate, avoiding a cold start where nothing is flagged until data exists. Ran RCA as a guarded LangGraph state machine rather than a free-form prompt chain. Typed tools, timeouts and schema validation around every model call keep reasoning auditable and isolate failures to a single node. That is what minimises hallucinated incident statements: the difference between a demo and something an SRE trusts at 3am. Anonymised PII at ingestion with Presidio + spaCy, not on read. Stripping sensitive data before it lands in storage or reaches the LLM is a stricter boundary than redacting at query time. The reprocessing overhead is far cheaper than the blast radius of a leak.

Project source
Outcome

OpsNexus

Research proposal with a scoped pilot. Decision-speed and analysis-effort improvements are targets, not measured results. Ranked #1 of all submissions in HPE's FOCUS AI Ideathon 2026 Advanced to final review before the FOCUS Group leadership panel Phase-1 pilot scoped to two systems and one workflow: no new data platform, no migration

Project source

Hexus

An internal observability platform with self-serve model workflows, anomaly detection, and automated RCA. Automated RCA + anomaly detection across multi-tenant log streams Tenant isolation designed around database Row-Level Security, with policy and privileged-access testing still required Automated PII detection and anonymization at ingestion; detection coverage still requires validation Self-serve ML lifecycle: train, vectorize and activate models via API

Project source
Pilot targets

OpsNexus

40% faster decisions and 60% less manual analysis, gated over an 8-week pilot Aimed at the ~65% of engineering effort HPE currently spends assembling context

Project source

Hexus

Not specified in this project record.

Evidence

OpsNexus

Architecture, engineering decisions, and the Phase-1 pilot plan are documented in this case study.

Project source

Hexus

The local RCA demo uses synthetic incidents, not employer telemetry. Component explanations and engineering decisions describe the platform workflow.

Project source
Portfolio reference

Capabilities and systems

Agentic AI

Reasoning becomes useful when context, tools and human decisions have explicit boundaries.

Backend Engineering

Contracts, tenant boundaries and data access shape the service, not just its endpoints.

Distributed Systems

Independent services trade coordination and operational cost for clearer boundaries and release freedom.

  • Decomposed the platform into 12+ focused microservices over a shared monolith.Enterprise Production
    Engineering detail

    Independent services let 24+ features ship on their own cadence to 500+ orgs without one release train blocking everything. The operational cost of running many services was absorbed by CI/CD automation.

    Enter this system
  • Distributed Software-Defined Network: overviewResearch
    Engineering detail

    A distributed SDN model with an accelerated Raft consensus algorithm that keeps multiple controllers consistent while scaling protocol handling. Top 8 of 200+ teams at the IEEE Kalpana 30Hr Hackathon.

  • AI Software Development Engineer / Hewlett Packard Enterprise (HPE)Aug 2025 – Present
    Engineering detail

    Ship production AI systems and the Kubernetes-native platform services beneath them for HPE's Emerging Technologies team. Architected HPE OpsNexus, an enterprise operations intelligence mesh that unifies siloed operational systems into a knowledge graph and reasons across domains to predict downstream business impact. It ranked #1 in HPE's FOCUS AI Ideathon 2026 and was taken to final review before the FOCUS Group leadership panel. Engineered 7 production-grade AI systems and the distributed backend services beneath them within HPE's Emerging Technologies Team, automating complex enterprise engineering workflows and giving engineers their time back. Architected 8 Kubernetes-native microservices and 16 modular AI orchestration services for scalable AI platforms, powering intelligent incident classification, retrieval-augmented generation, and the generation of 450+ standardized documents. Developed AI-powered workflow automation with human-in-the-loop (HITL) decision support wired into enterprise systems, reducing manual operational effort by 42%. Designed MLOps pipelines for multiple models across hybrid-cloud environments, alongside sustainable automation that improved infrastructure visibility and eliminated 14 engineering hours of manual analysis every week. Modernized legacy enterprise applications by migrating 16 backend services and their databases to containerized, cloud-native architectures, standardizing deployment, observability and rollback across the platform estate. Partnered with software, cloud and platform engineering teams across 4+ business units to deliver HPE OneAI, applying modern software engineering, DevOps and cloud-native practices. Contributing to Agentic & Augmented AI orchestration that integrates adaptive learning and real-time feedback (RLHF) for self-optimizing systems. Ramped rapidly on Red Hat OpenShift, OpenStack and HPE Ezmeral to support key enterprise accounts, sustaining platform stability through routine health checks, fast issue resolution, and clear operational SOPs (recognized in HPE's Points Recognition program). Endorsed by the Director of HPE Managed Services, with the recommendation published in HPE's magazine.

Cloud

The runtime, deployment process and operational handover are part of the software.

  • Cloud Engineering Intern / Hewlett Packard Enterprise (HPE)Feb 2025 – Jul 2025
    Engineering detail

    Automate secure, scalable infrastructure across AWS, Azure and HPE GreenLake. Engineered automation scripts and Ansible playbooks to streamline deployment pipelines and enhance infrastructure reliability across multi-cloud environments. Designed and tested cloud-native architectures with AWS, Azure, and HPE GreenLake, optimizing resource allocation and supporting scalable, elastic applications. Developed microservice packaging strategies with Docker and explored container orchestration using Kubernetes for modular application lifecycles. Configured secure Linux environments with RHEL system management, including ACL-based permissions, SELinux tuning, and network segmentation.

  • AWS / GTMScaleEnterprise Production
    Engineering detail

    AWS is part of GTMScale's technology stack.

    Enter this system
  • Derived design from live IaC and Kubernetes state, not the CMDB.Enterprise Internal
    Engineering detail

    ITSM approvals were being rejected for inaccurate topology. Reading the actual Terraform and Kubernetes manifests, the real source of truth, removed the drift that documentation-based mapping always accumulates.

    Enter this system
Kubernetes

Container orchestration and infrastructure topology connect application intent to running services.

  • Kubernetes / Hexus – Smart Log ManagerEnterprise Internal
    Engineering detail

    Kubernetes is part of Hexus – Smart Log Manager's technology stack.

    Enter this system
  • Agentic AIOps Blueprint: IaC / K8sEnterprise Internal
    Engineering detail

    Terraform and Kubernetes manifests supply resource definitions instead of manually maintained topology descriptions.

    Enter this system
  • Cloud Engineering Intern / Hewlett Packard Enterprise (HPE)Feb 2025 – Jul 2025
    Engineering detail

    Automate secure, scalable infrastructure across AWS, Azure and HPE GreenLake. Engineered automation scripts and Ansible playbooks to streamline deployment pipelines and enhance infrastructure reliability across multi-cloud environments. Designed and tested cloud-native architectures with AWS, Azure, and HPE GreenLake, optimizing resource allocation and supporting scalable, elastic applications. Developed microservice packaging strategies with Docker and explored container orchestration using Kubernetes for modular application lifecycles. Configured secure Linux environments with RHEL system management, including ACL-based permissions, SELinux tuning, and network segmentation.

Full Stack

An interface, API and data model should explain the same product.

  • GTMScale: overviewEnterprise Production
    Engineering detail

    A microservice-based B2B SaaS ERP (MERN + AWS) powering 500+ healthcare organizations, with 24+ features, real-time dashboards and sales trackers, hardened by JWT auth and automated testing across 750+ CI/CD deployments.

    Enter this system
  • Hexus – Smart Log Manager: Micro-FrontendEnterprise Internal
    Engineering detail

    The federated React frontend exposes model management and training observability.

    Enter this system
  • NetworkNex: overviewLive Deployment
    Engineering detail

    A LinkedIn-style app with API integration, a dynamic feed, dark mode and secure Google-based session management, built on Next.js, the MERN stack and Prisma.

  • Ticket Snatcher: overviewLive Deployment
    Engineering detail

    A MERN movie-booking app. Theatre and showtime filters narrow the catalogue to a single screening, an interactive seat map carries reservation state per seat, and an Express API over MongoDB persists the booking. Deployed on Vercel.

  • My Keeper: overviewLive Deployment
    Engineering detail

    A React note manager with full create, read, update and delete over component state, an expanding composer that stays collapsed until focused, and a responsive card board. Deployed on Vercel.

MLOps

Training, activation, evaluation and operating a model belong to one lifecycle.

  • AI Software Development Engineer / Hewlett Packard Enterprise (HPE)Aug 2025 – Present
    Engineering detail

    Ship production AI systems and the Kubernetes-native platform services beneath them for HPE's Emerging Technologies team. Architected HPE OpsNexus, an enterprise operations intelligence mesh that unifies siloed operational systems into a knowledge graph and reasons across domains to predict downstream business impact. It ranked #1 in HPE's FOCUS AI Ideathon 2026 and was taken to final review before the FOCUS Group leadership panel. Engineered 7 production-grade AI systems and the distributed backend services beneath them within HPE's Emerging Technologies Team, automating complex enterprise engineering workflows and giving engineers their time back. Architected 8 Kubernetes-native microservices and 16 modular AI orchestration services for scalable AI platforms, powering intelligent incident classification, retrieval-augmented generation, and the generation of 450+ standardized documents. Developed AI-powered workflow automation with human-in-the-loop (HITL) decision support wired into enterprise systems, reducing manual operational effort by 42%. Designed MLOps pipelines for multiple models across hybrid-cloud environments, alongside sustainable automation that improved infrastructure visibility and eliminated 14 engineering hours of manual analysis every week. Modernized legacy enterprise applications by migrating 16 backend services and their databases to containerized, cloud-native architectures, standardizing deployment, observability and rollback across the platform estate. Partnered with software, cloud and platform engineering teams across 4+ business units to deliver HPE OneAI, applying modern software engineering, DevOps and cloud-native practices. Contributing to Agentic & Augmented AI orchestration that integrates adaptive learning and real-time feedback (RLHF) for self-optimizing systems. Ramped rapidly on Red Hat OpenShift, OpenStack and HPE Ezmeral to support key enterprise accounts, sustaining platform stability through routine health checks, fast issue resolution, and clear operational SOPs (recognized in HPE's Points Recognition program). Endorsed by the Director of HPE Managed Services, with the recommendation published in HPE's magazine.

  • Hexus – Smart Log Manager: Anomaly MLEnterprise Internal
    Engineering detail

    Use Isolation Forest and LOF for unsupervised anomaly detection, with supervised scoring as labels accumulate.

    Enter this system
  • Multi-Modal AI Speech Coach: overviewLive Deployment
    Engineering detail

    An AI-driven app that personalises interactive coaching prompts with real-time LLM workflows, pairing TensorFlow speech analysis with a dynamic ReactJS UI. Top 6 at the Nex-Gen AI 24Hr Hackathon.

  • HRBIRS: overviewResearch
    Engineering detail

    Principal-Author research on a blockchain-based recruitment system using Ethereum + Solidity (PoS) for secure, immutable data, paired with a hybrid TF-IDF recommendation engine achieving 85% precision. Published at ICT4SD 2025.

Security

Identity, secrets, tenant isolation and redaction must protect the data before reasoning begins.

Automation

Make repeated engineering work reproducible, reviewable and connected to its source.

  • Wired generation into CI/CD rather than a manual 'generate docs' step.Production Pilot
    Engineering detail

    CI/CD makes regeneration part of a code change rather than a separate manual task. Failed jobs, uncovered modules and semantic inaccuracies can still leave gaps, so freshness checks and human review remain necessary.

    Enter this system
  • Emitted native, compliant Word / SOP artifacts in the exact format ITSM already requires.Enterprise Internal
    Engineering detail

    Compliance reviewers live in fixed Word templates. Generating those deliverables directly eliminated manual reformatting entirely, a larger operational win than asking an org to adopt a new tool.

    Enter this system
  • Cloud Engineering Intern / Hewlett Packard Enterprise (HPE)Feb 2025 – Jul 2025
    Engineering detail

    Automate secure, scalable infrastructure across AWS, Azure and HPE GreenLake. Engineered automation scripts and Ansible playbooks to streamline deployment pipelines and enhance infrastructure reliability across multi-cloud environments. Designed and tested cloud-native architectures with AWS, Azure, and HPE GreenLake, optimizing resource allocation and supporting scalable, elastic applications. Developed microservice packaging strategies with Docker and explored container orchestration using Kubernetes for modular application lifecycles. Configured secure Linux environments with RHEL system management, including ACL-based permissions, SELinux tuning, and network segmentation.

3D / Web Engineering

Make architecture inspectable while keeping the text, keyboard and mobile experience complete.

  • This portfolioInteractive implementation
    Engineering detail

    This portfolio combines server-rendered Next.js and React content with Three.js spatial architecture, instanced geometry, demand rendering, keyboard inspection and a contextual AI Twin.

Production AI / Portfolio synthesis

  1. User

    A bounded request enters the system. Intent is captured before any tool or model can act.

  2. Frontend

    Expose the request and its state. People need to see what the system accepted and what is still pending.

  3. API

    Check the contract and identity. Authentication and tenant policy are boundaries, not a prompt instruction.

  4. Intent

    Choose a supported workflow. A constrained plan prevents an open-ended request from becoming an unrestricted action.

  5. Orchestrator

    Coordinate reasoning and tool steps. State, typed tools and timeouts keep a failed step local.

  6. Retrieval

    Select relevant, permitted context. The proposed graph and retrieval path keeps relationships alongside documents.

  7. Model

    Synthesize over supplied context. Generated prose is separate from the facts and contracts that constrain it.

  8. Tool

    Execute a registered operation. A typed interface bounds what orchestration can request.

  9. Infrastructure

    Host the service and its dependencies. Application behavior also depends on the container and cloud runtime.

    • Cloud Engineering Intern / Hewlett Packard Enterprise (HPE)Feb 2025 – Jul 2025
      Engineering detail

      Automate secure, scalable infrastructure across AWS, Azure and HPE GreenLake. Engineered automation scripts and Ansible playbooks to streamline deployment pipelines and enhance infrastructure reliability across multi-cloud environments. Designed and tested cloud-native architectures with AWS, Azure, and HPE GreenLake, optimizing resource allocation and supporting scalable, elastic applications. Developed microservice packaging strategies with Docker and explored container orchestration using Kubernetes for modular application lifecycles. Configured secure Linux environments with RHEL system management, including ACL-based permissions, SELinux tuning, and network segmentation.

    • Docker / Hexus – Smart Log ManagerEnterprise Internal
      Engineering detail

      Docker is part of Hexus – Smart Log Manager's technology stack.

      Enter this system

  10. Validation

    Check the output against explicit criteria. A plausible response must still satisfy a schema and its evidence boundary.

  11. Human approval

    Pause at the decision boundary. The OpsNexus research design makes human approval an architectural gate.

  12. Action

    Apply only the approved operation. Approval and execution are separate states in this local synthesis.

    • AI Software Development Engineer / Hewlett Packard Enterprise (HPE)Aug 2025 – Present
      Engineering detail

      Ship production AI systems and the Kubernetes-native platform services beneath them for HPE's Emerging Technologies team. Architected HPE OpsNexus, an enterprise operations intelligence mesh that unifies siloed operational systems into a knowledge graph and reasons across domains to predict downstream business impact. It ranked #1 in HPE's FOCUS AI Ideathon 2026 and was taken to final review before the FOCUS Group leadership panel. Engineered 7 production-grade AI systems and the distributed backend services beneath them within HPE's Emerging Technologies Team, automating complex enterprise engineering workflows and giving engineers their time back. Architected 8 Kubernetes-native microservices and 16 modular AI orchestration services for scalable AI platforms, powering intelligent incident classification, retrieval-augmented generation, and the generation of 450+ standardized documents. Developed AI-powered workflow automation with human-in-the-loop (HITL) decision support wired into enterprise systems, reducing manual operational effort by 42%. Designed MLOps pipelines for multiple models across hybrid-cloud environments, alongside sustainable automation that improved infrastructure visibility and eliminated 14 engineering hours of manual analysis every week. Modernized legacy enterprise applications by migrating 16 backend services and their databases to containerized, cloud-native architectures, standardizing deployment, observability and rollback across the platform estate. Partnered with software, cloud and platform engineering teams across 4+ business units to deliver HPE OneAI, applying modern software engineering, DevOps and cloud-native practices. Contributing to Agentic & Augmented AI orchestration that integrates adaptive learning and real-time feedback (RLHF) for self-optimizing systems. Ramped rapidly on Red Hat OpenShift, OpenStack and HPE Ezmeral to support key enterprise accounts, sustaining platform stability through routine health checks, fast issue resolution, and clear operational SOPs (recognized in HPE's Points Recognition program). Endorsed by the Director of HPE Managed Services, with the recommendation published in HPE's magazine.

  13. Outcome

    Return an inspectable result. Outcome and overrides close the proposed operational-memory loop.

Distributed systems / Representative topology

  1. Gateway

    Authenticate and route incoming traffic. A clear API boundary connects clients to independent services.

  2. Services

    Own independent product responsibilities. GTMScale's service boundaries support separate release cadences.

  3. Cache

    An illustrative fast-read branch. This fixture can miss without stopping the primary data path. It is not a GTMScale architecture claim.

    • Database

      Persist operational and relational records. Query tuning, indexes and pagination control work before additional infrastructure.

    • Queue

      An illustrative asynchronous boundary. Pending work can wait while a worker recovers; this queue is a local simulation component.

      • Worker

        Consume one captured unit of work. The fixture retries a failed worker without replaying completed request stages.

        • External service

          A separately owned dependency. The OpsNexus design consumes published APIs rather than replacing systems of record.

        • Retry boundary

          Bound the recovery attempt. The representative worker branch stops after its configured retry budget.

        • Observability

          Expose service state and signals. The operator needs to distinguish an affected dependency from a healthy branch.

        • Delivery

          Test and release services. Release automation absorbs some of the operational cost of independent services.

        Agent system / Portfolio synthesis

        1. Planner

          Decompose a supported task. OpsNexus proposes durable reasoning steps, rather than one unbounded prompt.

        2. Context

          Carry extracted facts into reasoning. RapidDoc extracts source structure before language-model authoring.

        3. Retrieval

          Find supporting context. Graph traversal and document retrieval answer different parts of an operational question.

        4. Parallel agents

          Fan work out by responsibility. RapidDoc trades coordination complexity for parallel documentation work.

        5. Tools

          Constrain executable operations. Typed tools and timeouts place boundaries around the Hexus RCA workflow.

        6. Memory

          Connect prior decisions and outcomes. The proposed OpsNexus memory graph distinguishes operational history from transient context.

        7. Evaluation

          Inspect a result before it advances. Schema checks are a guard; they are not evidence that every generated statement is true.

        8. Human gate

          Approve or reject the proposed action. The OpsNexus design includes human approval as an explicit architectural component.

        9. Feedback

          Record the outcome and override. A feedback path should retain the context of the decision it follows.

        Engineering pipeline / Portfolio synthesis

        1. Idea

          Name the problem and the intended outcome. The project problem statements establish what each system was built to change.

          • GTMScale: problemEnterprise Production
            Engineering detail

            Healthcare organizations needed a reliable, fast, multi-tenant ERP for day-to-day operations and analytics, shipped continuously without breaking production.

            Enter this system

        2. Design

          Choose boundaries and trade-offs. Architecture decisions determine what can evolve independently.

        3. Code

          Implement behavior and extract structure. RapidDoc uses code-derived relationships as the input to documentation.

        4. Test

          Stop a broken change before release. GTMScale's automated test gates support continuous service delivery.

        5. Container

          Package the service with a runtime boundary. HPE cloud work includes Docker packaging and Kubernetes orchestration.

          • Cloud Engineering Intern / Hewlett Packard Enterprise (HPE)Feb 2025 – Jul 2025
            Engineering detail

            Automate secure, scalable infrastructure across AWS, Azure and HPE GreenLake. Engineered automation scripts and Ansible playbooks to streamline deployment pipelines and enhance infrastructure reliability across multi-cloud environments. Designed and tested cloud-native architectures with AWS, Azure, and HPE GreenLake, optimizing resource allocation and supporting scalable, elastic applications. Developed microservice packaging strategies with Docker and explored container orchestration using Kubernetes for modular application lifecycles. Configured secure Linux environments with RHEL system management, including ACL-based permissions, SELinux tuning, and network segmentation.

        6. Deploy

          Release the artifact and its handover material. AIOps connects infrastructure changes to operational design documents; GTMScale releases services.

        7. Observe

          Inspect behavior in its operating context. HPE platform work brings deployment, observability and rollback into the application lifecycle.

          • AI Software Development Engineer / Hewlett Packard Enterprise (HPE)Aug 2025 – Present
            Engineering detail

            Ship production AI systems and the Kubernetes-native platform services beneath them for HPE's Emerging Technologies team. Architected HPE OpsNexus, an enterprise operations intelligence mesh that unifies siloed operational systems into a knowledge graph and reasons across domains to predict downstream business impact. It ranked #1 in HPE's FOCUS AI Ideathon 2026 and was taken to final review before the FOCUS Group leadership panel. Engineered 7 production-grade AI systems and the distributed backend services beneath them within HPE's Emerging Technologies Team, automating complex enterprise engineering workflows and giving engineers their time back. Architected 8 Kubernetes-native microservices and 16 modular AI orchestration services for scalable AI platforms, powering intelligent incident classification, retrieval-augmented generation, and the generation of 450+ standardized documents. Developed AI-powered workflow automation with human-in-the-loop (HITL) decision support wired into enterprise systems, reducing manual operational effort by 42%. Designed MLOps pipelines for multiple models across hybrid-cloud environments, alongside sustainable automation that improved infrastructure visibility and eliminated 14 engineering hours of manual analysis every week. Modernized legacy enterprise applications by migrating 16 backend services and their databases to containerized, cloud-native architectures, standardizing deployment, observability and rollback across the platform estate. Partnered with software, cloud and platform engineering teams across 4+ business units to deliver HPE OneAI, applying modern software engineering, DevOps and cloud-native practices. Contributing to Agentic & Augmented AI orchestration that integrates adaptive learning and real-time feedback (RLHF) for self-optimizing systems. Ramped rapidly on Red Hat OpenShift, OpenStack and HPE Ezmeral to support key enterprise accounts, sustaining platform stability through routine health checks, fast issue resolution, and clear operational SOPs (recognized in HPE's Points Recognition program). Endorsed by the Director of HPE Managed Services, with the recommendation published in HPE's magazine.

        8. Improve

          Feed evidence into the next change. Query optimization and CI regeneration are concrete examples of iterative engineering.

        Project journeys

        Follow an incident

        A synthetic local fixture follows the Hexus incident workflow. It uses no employer telemetry and executes no infrastructure actions.

        1. Identify the signal

          A service and severity give triage an initial scope. The documented API coordinates ingestion; database policies, not this parser, enforce tenant isolation.

        2. Score the fixture

          Hexus describes unsupervised detection when labelled incidents are scarce, with supervised scoring added as labels accumulate.

        3. Inspect redaction

          The ingestion-redaction decision removes sensitive fields before persistence or model access, rather than relying on redaction at read time.

        4. Inspect supporting signals

          Context should remain inside tenant boundaries. The architecture documents API-to-agent and agent-to-database relationships, with RLS as a separate control.

        5. Explain the rule match

          The documented LangGraph decision uses typed tools, timeouts and schema validation around model calls. The local fixture demonstrates a narrower boundary: known rule or human triage.

        6. Review the next step

          An unmatched rule must remain unknown. The demo cannot confirm a root cause, approve a change, or act on infrastructure.

        Hexus references
        Follow a document

        A bounded browser parser and local templates produce this captured document. Source architecture links describe the project; no live agent, model, repository or ITSM system is called.

        1. Analyze the input

          Capture source text and locate definitions in the local fixture.

        2. Extract signatures

          Inspect parameters, types and return values extracted from the input.

        3. Inspect responsibility

          Name and body heuristics provide a bounded description for each symbol.

        4. Generate documentation

          Local language-aware templates turn the extracted records into docstrings or JSDoc.

        5. Assemble the artifact

          Combine function references and the component map into Markdown.

        6. Review CI-ready output

          Inspect the document before any repository integration.

        Project artifact
        Follow infrastructure into a document

        A bounded browser parser and local templates produce this captured document. Source architecture links describe the project; no live agent, model, repository or ITSM system is called.

        1. Read the manifest

          Capture the Kubernetes or Terraform fixture and identify resource blocks.

        2. Map relationships

          Inspect resource references and dependencies from the bounded parser.

        3. Assign responsibilities

          Group resources into compute, network, data, security and observability.

        4. Review local checks

          Inspect the fixture's policy findings, including exposure, limits and access hints.

        5. Assemble the LLD

          Build the local design sections from the extracted resource graph.

        6. Inspect the deliverable

          Review the generated design and operational handover content.

        Project artifact

        Request and deployment reference

        Representative local fixtures. An approval is a visitor decision inside the demonstration, not an infrastructure change.

        Follow a request

        1. User

          A bounded request enters the system. Intent is captured before any tool or model can act.

        2. Frontend

          Expose the request and its state. People need to see what the system accepted and what is still pending.

        3. API

          Check the contract and identity. Authentication and tenant policy are boundaries, not a prompt instruction.

        4. Intent

          Choose a supported workflow. A constrained plan prevents an open-ended request from becoming an unrestricted action.

        5. Orchestrator

          Coordinate reasoning and tool steps. State, typed tools and timeouts keep a failed step local.

        6. Retrieval

          Select relevant, permitted context. The proposed graph and retrieval path keeps relationships alongside documents.

        7. Model

          Synthesize over supplied context. Generated prose is separate from the facts and contracts that constrain it.

        8. Tool

          Execute a registered operation. A typed interface bounds what orchestration can request.

        9. Infrastructure

          Host the service and its dependencies. Application behavior also depends on the container and cloud runtime.

        10. Validation

          Check the output against explicit criteria. A plausible response must still satisfy a schema and its evidence boundary.

        11. Human approval

          Pause at the decision boundary. The OpsNexus research design makes human approval an architectural gate.

          An explicit approval is required to continue. Rejection performs no action.

        12. Action

          Apply only the approved operation. Approval and execution are separate states in this local synthesis.

        13. Outcome

          Return an inspectable result. Outcome and overrides close the proposed operational-memory loop.

        Follow a deployment

        1. Source

          Capture a candidate change. A release begins with a specific artifact, not a moving target.

        2. Build & test

          Validate the candidate before packaging. Automated test gates are part of GTMScale's documented release workflow.

        3. Container

          Package the application boundary. HPE cloud work includes Docker packaging and Kubernetes orchestration.

        4. Registry

          Resolve a versioned image. This registry is a representative local fixture, not a claim about an employer's registry.

        5. Cluster

          Prepare the workload topology. The representative rollout connects the image to service capacity; no cluster is contacted.

        6. Rollout review

          Hold traffic behind a visitor decision. This explicit approval is part of the local demonstration. AIOps separately documents ITSM handover artifacts.

          An explicit approval is required to continue. Rejection performs no action.

        7. Service

          Apply the approved local rollout state. The scenario changes only its in-memory service state.

        8. Traffic

          Trace a synthetic request through the service. Readiness and traffic are separate concerns in a release.

        9. Monitoring

          Inspect the fixture's resulting state. Observability is part of the documented HPE platform lifecycle, not a live feed on this page.

        10. Feedback

          Record the outcome and handover artifact. Source-linked operational documentation connects deployment and the next engineering change.

        Follow a background request

        1. Gateway

          Authenticate and route incoming traffic. A clear API boundary connects clients to independent services.

        2. Services

          Own independent product responsibilities. GTMScale's service boundaries support separate release cadences.

        3. Queue

          An illustrative asynchronous boundary. Pending work can wait while a worker recovers; this queue is a local simulation component.

        4. Worker

          Consume one captured unit of work. The fixture retries a failed worker without replaying completed request stages.

        5. External service

          A separately owned dependency. The OpsNexus design consumes published APIs rather than replacing systems of record.

        6. Database

          Persist operational and relational records. Query tuning, indexes and pagination control work before additional infrastructure.

        7. Observability

          Expose service state and signals. The operator needs to distinguish an affected dependency from a healthy branch.

        Follow agent execution

        1. Planner

          Decompose a supported task. OpsNexus proposes durable reasoning steps, rather than one unbounded prompt.

        2. Context

          Carry extracted facts into reasoning. RapidDoc extracts source structure before language-model authoring.

        3. Retrieval

          Find supporting context. Graph traversal and document retrieval answer different parts of an operational question.

        4. Parallel agents

          Fan work out by responsibility. RapidDoc trades coordination complexity for parallel documentation work.

        5. Tools

          Constrain executable operations. Typed tools and timeouts place boundaries around the Hexus RCA workflow.

        6. Evaluation

          Inspect a result before it advances. Schema checks are a guard; they are not evidence that every generated statement is true.

        7. Human gate

          Approve or reject the proposed action. The OpsNexus design includes human approval as an explicit architectural component.

          An explicit approval is required to continue. Rejection performs no action.

        8. Feedback

          Record the outcome and override. A feedback path should retain the context of the decision it follows.

        9. Memory

          Connect prior decisions and outcomes. The proposed OpsNexus memory graph distinguishes operational history from transient context.