Profile & Technical Leadership

Architecture Shaped by Operating Realities.

I work at the convergence of Agentic AI, high-throughput distributed systems, and stringent enterprise compliance—turning theoretical LLM capabilities into production platforms that teams can trust, evaluate, and scale.

Proven Track Record

8+ Years of Production Engineering

From Fortune 50 healthcare (Johnson & Johnson) and Dubai's largest digital marketplace to next-generation enterprise Agentic AI platforms.

Eva Tek

Agentic AI SaaS
Principal AI Systems Architect
Sep 2024 — Present
Remote, USA

Architect and deploy production-grade AI orchestration engine running stateful multi-agent workflows with LangGraph, GCP serverless, and Vertex AI.

Core Engineering Deliverables & Impact:
Engineered production-grade RAG architectures with FastAPI, Dataflow, BigQuery, hybrid vector search, and Redis-based state checkpointing.
Established LLMOps and observability frameworks with LangSmith and OpenTelemetry, reducing token consumption and average run latency by over 40%.
Engineered high-availability agent runtime on Kubernetes, maintaining 99%+ uptime via automated circuit breakers and local open-weight SLM fallback.
Enforced AI risk controls through strict Pydantic v2 schema validation for deterministic tool calling.
Technologies & Protocols Applied:
LangGraphModel Context Protocol (MCP)GCP Vertex AIRedis CheckpointingFastAPIOpenTelemetryKubernetesPydantic
Academic Education

University of Engineering & Technology (UET), Lahore

Bachelor of Science in Computer Science (2014 — 2018)

Rigorous foundation in Distributed Algorithms, Operating Systems, Data Structures & Machine Learning.

Verified Industry Certifications
Google Advanced Data Analytics ProfessionalGoogle / Coursera AWS Cloud Solutions ArchitectAmazon Web Services / Coursera Aha! Product Management Professional CertificateAha! / LinkedIn IBM Full Stack Software Developer Professional CertificateIBM / Coursera

1. Deterministic Over Chaotic

LLMs should never run free-form without typed schema contracts. Every agent step must be validated with strict Pydantic models and explicit state machine graphs.

2. Durable Over Ephemeral

Workflows fail when memory lives only in prompt context. Redis checkpointing provides instant session resurrection without burning thousands of wasted tokens.

3. Sovereign Over Dependent

Enterprises cannot risk upstream API outages or data leakage. Automated circuit breakers dynamically switch to self-hosted open-weight SLMs for 99%+ uptime.