AI Engineer | AI Automation & QA Engineer

I build AI automation, AI agents, and LLM pipelines that keep working when models, APIs, and data misbehave. Most of my work sits where AI meets backend and cloud engineering: n8n workflows, Python services, PostgreSQL state, and AWS. QA and reliability engineering is the thread through all of it. I test the systems I build the way I test other people's: at the boundaries, at the API, and on the failure paths.

Karachi, Pakistan. Open to remote work.

Portfolio · LinkedIn · Email · ORCID

• AI automation and agents: stateful agents, tool-using workflows, structured LLM output with validation, retries, and human escalation
• n8n and workflow automation: self-hosted n8n, PostgreSQL-backed workflow state, dead-letter queues, idempotent writes, Make.com
• LLM systems: multi-stage LLM pipelines, RAG and Corrective RAG, LangGraph, LangChain, Groq, Gemini, OpenRouter
• Backend and cloud: Python, FastAPI, PostgreSQL, Supabase, Docker, AWS (EC2, Lambda, EventBridge, SQS, RDS, S3), Terraform
• QA and reliability: B2B SaaS QA, API testing, multi-tenant isolation, RBAC, business-logic and financial-integrity testing

Production work as Backend & AI Automation Engineer. Client and internal details are not published.

• 28K+ production AI enrichment pipeline: status-driven PostgreSQL pipeline with 4 sequential LLM stages (including translation into 9 languages) over 28,000+ records; 5.6x processing speedup, as reported
• AWS infrastructure and cost optimization: monthly AWS costs for the same platform reduced by 34%
• Multi-tenant B2B SaaS QA: lead QA on PharmaConnect across 20 end-to-end journeys and 6 roles; 80+ defects documented with API-level evidence and 11 critical P1 findings
• 678+ defects identified and documented across all of my QA engagements

Sentinel-Mesh is a research framework for remediating Terraform cloud-security misconfigurations. An LLM proposes a candidate patch; a Z3 SMT verifier checks it against explicit security invariants defined in a Cloud Perimeter Model; a rejected patch's counterexample is fed back to the model for another attempt. The model only proposes. The verifier decides.

• CloudFix-Bench: 105 AWS Terraform misconfiguration cases, archived on Zenodo for reproducibility
• Results on that benchmark: 88/105 cases remediated (83.81%, 95% Wilson interval 75.59% to 89.64%), against 64.76% without counterexample feedback and 35.24% for a Checkov baseline
• No security regressions observed against the modeled invariants (0.0%, 95% interval 0.00% to 3.45%)
• Scope: formal guarantees hold only within the Cloud Perimeter Model. Properties outside it, such as WAF associations or logging policies, are checked by the verifier without a formal proof certificate, and an external set of 12 cases reached 6/12.

Preprint on Research Square, manuscript under review. Lead author.

Research Square preprint · CloudFix-Bench on Zenodo · Code · Technical article on Medium

Also: peer reviewer for IEEE Access (5 verified reviews on Web of Science). B.S. Cybersecurity, Sir Syed University of Engineering & Technology.

• n8n-production-resilience-patterns: retries, dead-letter queues, idempotency, replay, and explicit state for self-hosted n8n
• enterprise-multitenant-qa-matrix: test matrices and checklists for multi-tenant isolation, RBAC, APIs, and business logic

I take on remote projects in:

• AI automation and API integrations
• n8n workflows (cloud or self-hosted)
• AI agents and LLM pipelines
• B2B SaaS QA, API testing, and multi-tenant testing

Portfolio · Upwork · Fiverr · LinkedIn · hira229922@gmail.com