Shubham – Python, LLM, AWS

Major Australian mining and resources companies needed faster responses from enterprise LLM applications before wider adoption. Shubham led technical presales and solution architecture, built customer POCs and deployed Redis Cloud as a central caching service.

• Led three customer engagements since August 2026.
• Created customer POCs and implemented Redis Cloud as a central caching service.
• Owned technical architecture and stakeholder engagement from the initial discussions through deployment, working with technical teams, directors and senior management.
• Result: Reduced total LLM response time by over 90% for supported workloads and supported a Redis contract representing $1M+ in potential revenue. The latency result was measured at the application response level.

Engineering teams needed a consistent way to use AI coding agents within enterprise architecture and security standards. Shubham built a developer harness that encoded approved software patterns and integrated them into GitHub-based development and deployment workflows.

• Designed and implemented the harness in Python, TypeScript and Node.js, with OpenAI and GitHub integration.
• Encoded reusable enterprise patterns and implemented feedback loops to refine the guidance from delivery outcomes.
• Result: Reduced token consumption by approximately 30% and rolled the harness out to the internal developer community. Adoption feedback was positive.

Engineering teams across a large multi-account AWS environment needed a repeatable way to move open-source models from experimentation into governed inference services. Shubham built a GPU-backed model-serving platform on Amazon SageMaker with automated Hugging Face model packaging and reusable CloudFormation templates.

• Implemented GPU-backed SageMaker inference for embedding and generative AI workloads, automated model packaging and deployment, and reusable CloudFormation infrastructure templates.
• Combined central governance with deployment into individual business-team accounts.
• Result: Established consistent serving patterns across 100+ AWS accounts and reduced manual packaging and infrastructure setup for teams moving models into managed inference. The platform standardised deployment, networking, access control and model lifecycle management.

Business applications needed access to enterprise knowledge and reusable AI capabilities without each team rebuilding retrieval and agent infrastructure. Shubham architected a shared GenAI platform with ingestion and embedding pipelines, hybrid vector and keyword retrieval, reranking and LangGraph agent orchestration.

• Architected the shared platform in Python and TypeScript on AWS.
• Built ingestion and embedding pipelines, hybrid vector/keyword retrieval with reranking, and ReAct-based agents in LangGraph that retrieved context and invoked enterprise tools.
• Result: Processed 500GB+ of enterprise content and provided reusable retrieval, agent and governance building blocks to multiple business applications, avoiding repeated implementation of these components by each team.

Engineering teams needed more efficient delivery pipelines, consistent compliance checks and reusable cloud infrastructure. Across successive engineering and platform roles, Shubham redesigned CI/CD with GitHub OIDC, introduced compliance checks across approximately 200 applications.

• Redesigned CI/CD using GitHub OIDC, introduced compliance checks across approximately 200 applications, and led five platform developers building reusable AWS patterns for Lambda, ECS and EKS.
• Result: Reduced monthly CI usage from approximately 100,000 to 50,000 minutes across around 100 teams. The GitHub OIDC product also reduced lead time for changes by 30%.
• Production operations: Helped achieve 99.9% platform uptime and reduced mean time to resolution by 60% through CloudWatch monitoring, alerting and Lambda incident runbooks.

A production AI assistant that helps procurement teams get accurate answers from large volumes of internal documentation and enterprise data. It combines RAG, agentic workflows, and vector search with enterprise integrations to reduce reliance on manual support channels.

• Ran discovery workshops with procurement stakeholders who had low AI literacy, clarifying security, scale, and cost requirements before scoping the build.
• Architected and productionised the RAG application with LangGraph, ChatKit, and PostgreSQL/pgvector.
• Designed retrieval and grounding workflows for complex procurement documents and knowledge sources.
• Integrated MCP-based enterprise services so agents could pull context from and act on internal systems.
• Implemented orchestration, prompt management, retrieval logic, and guardrails for enterprise production use.
• Reduced inbound procurement support queries by ~60% after adoption.