Umad Akram

Worked on infrastructure for Shopify Installments within the Money Movement & Expansion organization, building backend systems that supported high-volume payment and checkout workflows at scale. Focused on reliability, distributed systems, and internal tooling for improving developer productivity across payment services.

• Engineered infrastructure in Python powering Shopify Installments, supporting 10M+ daily payment events and improving system reliability across distributed payment services.
• Engineered an LLM-powered ML pipeline that automatically identified root causes behind failing end-to-end tests and executed remediation workflows, resolving 15k+ flaky tests weekly.
• Built a high-throughput API orchestration service integrating Shop Pay Installments with Affirm financing systems using Kubernetes, powering real-time installment package retrieval across 1M+ daily checkout requests.

Creation is a startup focused on automating the CAD design pipeline through AI-powered workflows inside SolidWorks. Worked on backend infrastructure, retrieval systems, and prompt-to-CAD generation pipelines that transformed natural language prompts into manufacturable 3D models.

• Developed a Python backend deployed on AWS that translated natural-language prompts into parametric CAD workflows, generating production-ready models and attracting a waitlist of 300+ engineers.
• Built an NLP pipeline using Claude API to parse unstructured design prompts and generate manufacturable CAD models through the SolidWorks API, supporting 100+ automated design generations.
• Engineered a RAG pipeline with 1,000+ PostgreSQL records to improve model accuracy, consistency, and iteration speed by retrieving past successful responses.

Worked on infrastructure automation, LLM systems, and internal platform tooling across enterprise banking systems. Focused on scalable backend infrastructure, Kubernetes automation, and observability for production AI services.

• Automated Kubernetes infrastructure provisioning using Go and Terraform across 150+ microservices, reducing environment setup time from 5 days to under 30 minutes.
• Designed a scalable LLM serving layer using FastAPI, Redis caching, and Kubernetes autoscaling to handle 25K+ daily prompt requests, reducing average response latency by 40ms.
• Implemented prompt logging, schema validation, and evaluation dashboards using OpenTelemetry and PostgreSQL, enabling structured analysis of hallucination patterns across 150K+ weekly LLM responses.