SHAFI ANWAR

I turn operational SOPs into production AI workflows.

I translate operational problems into AI systems.

My work starts with understanding how teams operate, where bottlenecks exist, and which decisions can be automated safely. From there, I design workflows that combine AI agents, APIs, business rules, and human review to create practical automation that works in the real world.

Currently, I contribute to automation initiatives at AWS while building production-focused AI projects around retrieval, workflow orchestration, and customer support automation.

Core stack

AWS Bedrock AgentsPythonYAMLAmazon S3Amazon QIntent Classification & RoutingMulti-step Agent ArchitectureStateful Workflow DesignHuman-in-the-Loop SystemsLow-Level DesignBusiness Process AutomationAPI IntegrationEvent-Driven ArchitectureAsync Workflow OrchestrationMulti-Turn Agent DesignConditional RoutingEscape Hatch PatternsJSON Schema DesignCase State ManagementPrompt EngineeringAmazon Bedrock (Claude)Streamlit

Production Credentials

Built internally at AWS

Reseller Support Micro-Agent

running in production

Converted a 30+ page SOP with dozens of branching decision paths into an AI agent that understands customer intent, applies conditional routing logic across linkage timing, billing, and solution provider restriction scenarios, and delivers the correct resolution to engineers in seconds. No manual SOP reading required.

500+

weekly queries

-20%

manual workload reduction

PythonGenAI Agent DevelopmentPrompt EngineeringREST API IntegrationJSON Schema DesignConditional Routing

Built internally at AWS

Route 53 SQIR Automation

testing - pre-production

End-to-end async agent that autonomously handles Route 53 service quota increase requests. Validates eligibility, obtains customer consent, creates service team tickets, locks the case, monitors resolution via event streams with a 96-hour escape hatch, and auto-resumes to deliver the final answer. Owned from architecture and LLD through to code and testing. Part of a larger SQIR initiative across 10+ AWS services.

10–15 min

manual process → zero human intervention

10+

AWS services in scope (DynamoDB, RDS, CloudFront, SNS, S3 and more)

PythonEvent-Driven ArchitectureMulti-Turn Async OrchestrationEventBridge-pattern TimersREST API IntegrationCase State ManagementGenAI Agent Development

Built internally at AWS

API Catalyst

poc - evaluated for adoption

AI-powered discovery tool that takes a natural language query from a microagent builder, searches 2,861+ onboarded AWS APIs, returns relevant matches for their specific use case, and verifies real-time platform onboarding status. Built a working prototype in 7 days, presented at the IN Together Summit Hackathon (April 2026), demoed to senior TPM leadership.

2,861+

APIs searchable

7 days

POC to live demo

PythonAmazon Bedrock (Claude)StreamlitNatural Language UnderstandingJSON CatalogREST API Integration

Portfolio Projects

Built in public. Designed to scale.

Self-directed production-grade builds. Each project solves a real problem and is designed to be extended into a product.

Nexatel AI Support Agent

pending bedrock access

A production-grade GCC telecom support agent built on AWS Bedrock Agents. Four IAM-isolated action groups, RAG over S3 Vectors, Bedrock Guardrails, API Gateway with WAF, and full CloudWatch observability. Navigated a regional infrastructure failure, two simultaneous AWS account verification gates, and a compliance correction mid-build. All documented transparently.

AWS Bedrock AgentsS3 VectorsBedrock GuardrailsLambdaDynamoDBAPI GatewayWAFCloudFrontCloudWatchPython
Open case file

More projects being added continuously. Next up: WPS Payroll Compliance Tool, Agentic RFP Generator, AWS Cost Anomaly Explainer.

Flagship Project

Nexatel - a GCC telecom support agent, built to show the whole architecture, including where it broke.

A self-directed portfolio project on AWS Bedrock: four IAM-isolated action groups, a retrieval-augmented knowledge base, guardrails, and a full production-style edge. Plus the real story - a regional infrastructure failure, two platform verification gates, and a compliance correction, all handled mid-build.

Open case file