Skip to content
LATEST: Beyond Benchmarks: Architecting Real-Time LLM Evaluation with Jev and LayaRead
SASHI KUMAR A

Principal Enterprise AI Architect

Sashi Kumar A

BuildingProduction-GradeAgentic Systems

Agentic Platforms · Multi-Agent Systems · MCP · Context Engineering · Distributed Systems · AI Governance

sashi_ai_architect.sh — expanded

# interactive — type help at the prompt below, or just start typing

$ uptime

20:27 up 18y, load average: 18 clients, 7 industries, 63+ agents

$ htop --impact

INFRA_COST [░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░████████████████████████████████████████████████████████████] -60%

DELIVERY_TIME [░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░████████████████████████████████████████████████████████████] -70%

MANUAL_TEST_EFFORT [░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░████████████████████████████████████████████████████████████] -40–70% ◌ target

DATA_ENG_TOIL [░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░████████████████████████████████████████████████████████████] -40–60% ◌ target

ANALYTICS_TURNAROUND hours [──────────────────────────────●] minutes

Tasks: 10 projects · 18 clients · 7 industries ◌ = stated target, not a measured result

$ git log --oneline --decorate career

  1. * feff02a (HEAD -> main, tag: 2022–present) feat(hi-tech): Agentic Data Operations Platform & AI Governance [+]

    Principal AI / Solutions Architect & Technical Delivery Lead · Global Engineering Services Firm · Fortune 10 · Technology

    • Led discovery, solution architecture and delivery planning for an enterprise platform automating data-product onboarding, engineering, governance and operations across Snowflake and AWS — targeting a 40–60% reduction in repetitive data-engineering work.
    • Architected a multi-agent system spanning metadata discovery, schema profiling, repository intelligence, transformation generation, ontology, orchestration and governance agents on Bedrock AgentCore and MCP.
    • Designed a versioned lineage architecture — canonical production graphs plus candidate graphs per feature branch — with automated impact analysis and human approval before promotion.
    • Designed CI/CD for dbt and Airflow with automated validation, lineage impact checks and approval gates on every merge request.
    • Defined enterprise controls: AgentCore Identity and Gateway, tool-level authorization, secrets, PII protection, prompt-injection defense and MCP security.
    open in WORK_EXPERIENCE ↗
  2. * 061623f feat(hi-tech): Agentic Software Quality Platform [+]

    Principal AI / Solutions Architect & Technical Delivery Lead · Global Engineering Services Firm · Fortune 10 · Technology

    • Led architecture of an AI-assisted software-quality platform that maps requirements to tests, generates and runs Playwright tests, and reports results through CI quality gates — targeting a 40–70% reduction in manual test-design and execution effort.
    • Designed the end-to-end workflow — code analysis, requirements-to-test mapping, generation, execution and quality-gate results returned to developers — on MCP-based tool integration.
    • Integrated multi-role authentication, screenshot validation, issue-tracker integration, end-to-end, performance/stress and P99 testing into the pipeline.
    • Shaped a phased roadmap through traceability to Human-in-the-Loop-approved updates of engineering documentation.
    open in WORK_EXPERIENCE ↗
  3. * f10a78b feat(hi-tech): Intelligent Document Processing & DMS Modernization [+]

    Principal AI / Solutions Architect & Technical Delivery Lead · Global Engineering Services Firm · Fortune 10 · Technology

    • Led solution architecture for an IDP platform integrated with an existing document-management estate of ~424K documents and multiple downstream applications.
    • Designed ingestion, classification, OCR/extraction, enrichment, confidence-based decisioning and Human-in-the-Loop validation for PDFs, structured documents and future CAD content.
    • Defined persona-aware workflows, enterprise SSO/LDAP integration, PII controls, document-version comparison, auditability and API/MCP downstream integrations.
    • Defined CI/CD and environment promotion with extraction-accuracy regression checks and controlled rollout of model-configuration changes.
    open in WORK_EXPERIENCE ↗
  4. * db24770 feat(aerospace): Secure Enterprise Knowledge Platform [+]

    Principal AI / Solutions Architect & Technical Delivery Lead · Global Engineering Services Firm · Fortune 50 · Aerospace

    • Led architecture of a secure GenAI knowledge platform giving engineering teams intelligent access to millions of export-controlled technical documents.
    • Integrated Amazon Kendra with fine-grained access control as the secure retrieval layer, and designed the evolution to Bedrock conversational search grounded only in authorized content.
    • Led CI/CD architecture for application services and cloud infrastructure, aligned with enterprise security and document-access requirements.
    open in WORK_EXPERIENCE ↗
  5. * 0e1e5cc feat(hi-tech): Enterprise Agentic Analytics Platform [+]

    Principal AI / Solutions Architect & Technical Delivery Lead · Global Engineering Services Firm · Digital Analytics Platform

    • Architected a natural-language-to-SQL analytics platform over large-scale Redshift data, cutting supported analytics turnaround from hours to minutes; designed for 600+ concurrent users.
    • Designed Supervisor, Metadata, Query Execution and Visualization agents on Bedrock AgentCore Runtime, Gateway, Identity, Memory and Observability.
    • Built on Redshift Data API, Lambda, DynamoDB, OpenSearch, SSE streaming and a React/TypeScript front end; took it from POC through architecture and SOW toward enterprise delivery.
    open in WORK_EXPERIENCE ↗
  6. * 3ef2ab2 feat(healthcare): Electronic Consent Management [+]

    Principal AI / Solutions Architect & Technical Delivery Lead · Global Engineering Services Firm · State Healthcare Agency

    • Designed and enhanced OCR/document extraction for healthcare provider forms, onboarding new providers and improving custom document adapters.
    • Led SOC 2 Type II and HIPAA compliance activities, including control evidence and audit readiness.
    • Designed a centralized AWS delivery pipeline (CodePipeline, CodeCommit, CodeBuild, CodeDeploy) integrated with IAM, KMS, Systems Manager and S3 for regulated workloads.
    open in WORK_EXPERIENCE ↗
  7. * 44af166 feat(public services): Application CI/CD Architecture [+]

    Principal AI / Solutions Architect & Technical Delivery Lead · Global Engineering Services Firm · Employee Benefits Provider

    • Led CI/CD architecture and reusable deployment patterns that reduced reliance on manual releases.
    • Guided teams on integrating delivery with infrastructure provisioning, access control and configuration management, and led deployment-readiness reviews.
    open in WORK_EXPERIENCE ↗
  8. * 8f11ace (tag: 2021–2022) feat(multi-sector): Enterprise Solutioning & Deal Shaping [+]

    Solutions Architect & Technical Pre-Sales · Technology Services Provider · Singapore · Multiple Enterprise Clients

    • Led solution discovery, architecture workshops, proposals, effort estimation and positioning across cloud, data and digital-transformation opportunities.
    • Translated customer requirements into solution architectures, SOW inputs, roadmaps, cost models and RFP/RFI responses with practice, sales and delivery leads.
    • Acted as customer-facing technical advisor — architecture reviews, risk and dependency analysis, executive presentations — and built CI/CD, IaC and release automation into every proposal.
    open in WORK_EXPERIENCE ↗
  9. * 71ce8e7 (tag: 2015–2020) feat(banking): Enterprise Cloud Modernization & Financial Risk Data Mart [+]

    Solutions Architect & Technical Delivery Lead · IT Consultancy · Singapore · Tier-1 Asian Bank

    • Architected large-scale AWS platforms for mission-critical banking workloads — high availability, fault tolerance, security and automated deployment.
    • Drove modernization programs contributing to ~60% infrastructure cost reduction and ~30% faster projected delivery.
    • Migrated financial-risk workloads from on-premises Oracle to RDS and Redshift; engineered ingestion of 150–200 GB of financial data daily.
    • Designed CI/CD for containerized and serverless banking services with repeatable, auditable releases.
    open in WORK_EXPERIENCE ↗
  10. * 732411f (tag: 2008–2015) feat(telecom): Enterprise Data Migration & Database Engineering [+]

    Solutions Architect & Technical Delivery Lead · Global IT Services Firm · Telecom & Healthcare Enterprises

    • Delivered data migration, integration and database-engineering programs across telecom, healthcare and analytics engagements.
    • Built and optimized high-volume Oracle/PL-SQL transformation and migration pipelines for large production rollouts.
    • Introduced version-controlled, scripted deployment of schema changes and PL/SQL releases across test and production.
    open in WORK_EXPERIENCE ↗

$ career --heatmap 2008..2026

'08 '09 '10 '11 '12 '13 '14 '15 '16 '17 '18 '19 '20 '21 '22 '23 '24 '25 '26

hi-tech · · · · · · · · · · · · · · · · ■■■ ■■■ ■■■

aerospace · · · · · · · · · · · · · · · · ■■■ ■■■ ■■■

healthcare ■■■ ■■■ ■■■ ■■■ ■■■ ■■■ ■■■ ■■■ · · · · · · ■■■ ■■■ ■■■ ■■■ ·

banking · · · · · · ■■■ ■■■ ■■■ ■■■ ■■■ ■■■ ■■■ ■■■ · · · · ·

telecom ■■■ ■■■ ■■■ ■■■ ■■■ ■■■ ■■■ ■■■ · · · · · · · · · · ·

public services · · · · · · · · · · · · · · ■■■ ■■■ ■■■ ■■■ ■■■

multi-sector · · · · · · · · · · · · · · ■■■ ■■■ ■■■ ■■■ ■■■

■■■ client work ■■■ multi-sector work hover a square for the project

$ kubectl get agents --all-namespaces

NAMESPACE NAME ROLE STATUS

analytics supervisor supervisor Running

analytics metadata specialist Running

analytics query-execution executor Running

analytics visualization specialist Running

data-ops metadata specialist Running

data-ops schema-profiling specialist Running

data-ops github specialist Running

data-ops transformation executor Running

data-ops ontology specialist Running

data-ops orchestration planner Running

data-ops governance verifier Running

cad-intel standards-compliance verifier Building

cad-intel revision-intelligence specialist Building

cad-intel search-knowledge router Building

cad-intel asset-space-intelligence specialist Building

… 48 more across other platforms (63+ total) · hover a row for its purpose

help · tab completes · ↑↓ history · esc closes

  • Agentic AI
  • Multi-Agent Systems
  • MCP
  • Bedrock AgentCore
  • Context Engineering
  • AI Security
  • Knowledge Systems
  • Data Platforms

01. FEATURED_ARCHITECTURES

Enterprise systems I have designed — the patterns, boundaries and production considerations behind them.

VIEW_ALL_ARCHITECTURES

Agentic Data Operations Platform

  1. Code
  2. Knowledge
  3. Graph
  4. Agents
  5. Validation
  6. Deployment

AI-native data engineering with lineage-aware automation, governed transformation and human-controlled deployment.

Agentic Data Engineering · Knowledge Graphs · Lineage Intelligence · Metadata AI

Explore Architecture

Agentic Software Intelligence

  1. Requirements
  2. Code
  3. Tests
  4. Evidence

Connecting requirements, code, tests and evidence through AI-driven software lifecycle intelligence.

Software Agents · Requirements Intelligence · Traceability · Test Generation

Explore Architecture

Secure Aerospace Knowledge Center

  1. Identity
  2. Authorization
  3. Knowledge
  4. Retrieval
  5. AI

Identity-aware engineering knowledge retrieval across millions of controlled technical documents.

Enterprise RAG · Fine-Grained Authorization · Engineering Knowledge · Hybrid Search

Explore Architecture

Enterprise Document Intelligence

  1. Documents
  2. Understanding
  3. Validation
  4. Knowledge

Transforming complex enterprise documents into structured, searchable and decision-ready knowledge.

Intelligent Document Processing · Multimodal AI · Semantic Extraction · Confidence Routing

Explore Architecture

Enterprise Agent Security

  1. Identity
  2. Context
  3. Tools
  4. Memory
  5. Execution

Defense-in-depth controls across identity, context, tools, memory, data and autonomous execution.

Agent Security · Prompt Injection Defense · Tool Authorization · Memory Security

Explore Architecture

02. FEATURED_THINKING

Deep dives into the parts of enterprise AI that only become interesting in production.

ALL_WRITING
Generative AI7 min

Beyond Benchmarks: Architecting Real-Time LLM Evaluation with Jev and Laya

Why production LLM reliability depends on a live evaluation pipeline rather than a benchmark score. Verify critical claims before display, while Jev or Laya triages sampled answers for deeper review and tested failover responds to sustained quality problems.

  • llm-evaluation
  • production-reliability
  • llm-as-judge
AI Agents9 min

Where Laya or Jev Fits in the Software Development Lifecycle

A practical architecture for using Jev or Laya to classify, route, and escalate software delivery work while keeping code generation, policy enforcement, and consequential approvals in the right hands.

  • agentic-ai
  • sdlc
  • jev

03. WORK_EXPERIENCE

17+ years across software engineering, cloud architecture and technical leadership. Clients are described by sector and scale — the names are confidential.

  1. PRINCIPAL_AI_SOLUTIONS_ARCHITECT_AND_TECHNICAL_DELIVERY_LEAD

    Global Engineering Services Firm

    2022 – Present

    Agentic Data Operations Platform & AI Governance

    Fortune 10 · Technology
    • Led discovery, solution architecture and delivery planning for an enterprise platform automating data-product onboarding, engineering, governance and operations across Snowflake and AWS — targeting a 40–60% reduction in repetitive data-engineering work.
    • Architected a multi-agent system spanning metadata discovery, schema profiling, repository intelligence, transformation generation, ontology, orchestration and governance agents on Bedrock AgentCore and MCP.
    • Designed a versioned lineage architecture — canonical production graphs plus candidate graphs per feature branch — with automated impact analysis and human approval before promotion.
    • Designed CI/CD for dbt and Airflow with automated validation, lineage impact checks and approval gates on every merge request.
    • Defined enterprise controls: AgentCore Identity and Gateway, tool-level authorization, secrets, PII protection, prompt-injection defense and MCP security.

    Agentic Software Quality Platform

    Fortune 10 · Technology
    • Led architecture of an AI-assisted software-quality platform that maps requirements to tests, generates and runs Playwright tests, and reports results through CI quality gates — targeting a 40–70% reduction in manual test-design and execution effort.
    • Designed the end-to-end workflow — code analysis, requirements-to-test mapping, generation, execution and quality-gate results returned to developers — on MCP-based tool integration.
    • Integrated multi-role authentication, screenshot validation, issue-tracker integration, end-to-end, performance/stress and P99 testing into the pipeline.
    • Shaped a phased roadmap through traceability to Human-in-the-Loop-approved updates of engineering documentation.

    Intelligent Document Processing & DMS Modernization

    Fortune 10 · Technology
    • Led solution architecture for an IDP platform integrated with an existing document-management estate of ~424K documents and multiple downstream applications.
    • Designed ingestion, classification, OCR/extraction, enrichment, confidence-based decisioning and Human-in-the-Loop validation for PDFs, structured documents and future CAD content.
    • Defined persona-aware workflows, enterprise SSO/LDAP integration, PII controls, document-version comparison, auditability and API/MCP downstream integrations.
    • Defined CI/CD and environment promotion with extraction-accuracy regression checks and controlled rollout of model-configuration changes.

    Secure Enterprise Knowledge Platform

    Fortune 50 · Aerospace
    • Led architecture of a secure GenAI knowledge platform giving engineering teams intelligent access to millions of export-controlled technical documents.
    • Integrated Amazon Kendra with fine-grained access control as the secure retrieval layer, and designed the evolution to Bedrock conversational search grounded only in authorized content.
    • Led CI/CD architecture for application services and cloud infrastructure, aligned with enterprise security and document-access requirements.

    Enterprise Agentic Analytics Platform

    Digital Analytics Platform
    • Architected a natural-language-to-SQL analytics platform over large-scale Redshift data, cutting supported analytics turnaround from hours to minutes; designed for 600+ concurrent users.
    • Designed Supervisor, Metadata, Query Execution and Visualization agents on Bedrock AgentCore Runtime, Gateway, Identity, Memory and Observability.
    • Built on Redshift Data API, Lambda, DynamoDB, OpenSearch, SSE streaming and a React/TypeScript front end; took it from POC through architecture and SOW toward enterprise delivery.

    Application CI/CD Architecture

    Employee Benefits Provider
    • Led CI/CD architecture and reusable deployment patterns that reduced reliance on manual releases.
    • Guided teams on integrating delivery with infrastructure provisioning, access control and configuration management, and led deployment-readiness reviews.
  2. SOLUTIONS_ARCHITECT_AND_TECHNICAL_PRE_SALES

    Technology Services Provider · Singapore

    2021 – 2022

    Enterprise Solutioning & Deal Shaping

    Multiple Enterprise Clients
    • Led solution discovery, architecture workshops, proposals, effort estimation and positioning across cloud, data and digital-transformation opportunities.
    • Translated customer requirements into solution architectures, SOW inputs, roadmaps, cost models and RFP/RFI responses with practice, sales and delivery leads.
    • Acted as customer-facing technical advisor — architecture reviews, risk and dependency analysis, executive presentations — and built CI/CD, IaC and release automation into every proposal.
  3. SOLUTIONS_ARCHITECT_AND_TECHNICAL_DELIVERY_LEAD

    IT Consultancy · Singapore

    2015 – 2020

    Enterprise Cloud Modernization & Financial Risk Data Mart

    Tier-1 Asian Bank
    • Architected large-scale AWS platforms for mission-critical banking workloads — high availability, fault tolerance, security and automated deployment.
    • Drove modernization programs contributing to ~60% infrastructure cost reduction and ~30% faster projected delivery.
    • Migrated financial-risk workloads from on-premises Oracle to RDS and Redshift; engineered ingestion of 150–200 GB of financial data daily.
    • Designed CI/CD for containerized and serverless banking services with repeatable, auditable releases.
  4. SOLUTIONS_ARCHITECT_AND_TECHNICAL_DELIVERY_LEAD

    Global IT Services Firm

    2008 – 2015

    Enterprise Data Migration & Database Engineering

    Telecom & Healthcare Enterprises
    • Delivered data migration, integration and database-engineering programs across telecom, healthcare and analytics engagements.
    • Built and optimized high-volume Oracle/PL-SQL transformation and migration pipelines for large production rollouts.
    • Introduced version-controlled, scripted deployment of schema changes and PL/SQL releases across test and production.

04. TECHNICAL_SKILLS

AI_ASSISTED_ENGINEERING

  • Claude
  • Amazon Bedrock
  • Bedrock AgentCore
  • Strands Agents
  • MCP
  • RAG
  • LLM evaluation
  • Prompt-injection defense
  • Tool security

PROGRAMMING_AND_AUTOMATION

  • Python
  • Go
  • TypeScript
  • Java
  • SQL
  • Scala
  • PL/SQL

CI_CD_AND_DEVELOPER_TOOLING

  • Jenkins
  • GitHub Actions
  • CodePipeline
  • CodeBuild
  • CodeDeploy
  • CI/CD quality gates

BUILD_TESTING_AND_QUALITY

  • Playwright
  • End-to-end testing
  • Performance / stress testing
  • P99 validation
  • Requirements-to-test traceability

INFRASTRUCTURE_AND_ENVIRONMENTS

  • Terraform
  • CloudFormation
  • Ansible
  • ECS
  • EC2
  • Lambda
  • Step Functions
  • API Gateway
  • S3
  • DynamoDB
  • RDS / Aurora
  • SQS / SNS
  • Kinesis

RELIABILITY_AND_OBSERVABILITY

  • CloudWatch
  • X-Ray
  • Langfuse
  • AgentCore Observability
  • High availability
  • Event-driven & distributed systems

DATA_PLATFORMS

  • Snowflake
  • Redshift
  • dbt
  • Airflow
  • AWS Glue
  • Spark / PySpark
  • PostgreSQL
  • Oracle
  • SQL Server
  • MySQL

05. ARCHITECTURE_LAB

Working explorations of enterprise AI architecture patterns — the implementations behind the writing.

  1. 001

    MCP Server with Bedrock AgentCore

    • Identity
    • Tool authorization
    • Gateway
    • Agent runtime
    In progress
  2. 002

    Version-Aware Enterprise RAG

    • Revision filtering
    • Entitlement-scoped retrieval
    • Context assembly
    In progress
  3. 003

    Multi-Tenant Agent Rate Limiting

    • Tenant identity
    • Token reservation
    • Cost attribution
    In progress
  4. 004

    Agent Memory Poisoning Defenses

    • Write-path validation
    • Provenance
    • Eviction under pressure
    In progress
  5. 005

    GitHub-to-Knowledge-Graph Agent

    • Repository parsing
    • Entity extraction
    • Lineage construction
    In progress

06. GET_IN_TOUCH

I work at the intersection of AI systems, cloud architecture and data platforms.

My focus is not getting an LLM demo working. It is the identity, context, memory, security, governance, observability and distributed-system concerns required to operate AI systems in production. If you are taking agentic AI from prototype to something a business can depend on, I'd like to hear about it.

CONNECT_WITH_ME