Luis Norman

Software Development Engineer II · Core Services Search

• Redesigned Amazon Music's search consistency after finding the read path was silently dropping stale items when catalog notifications got lost. Built an async reconciliation pipeline that detects divergences and triggers targeted reindexing. Cut p99 20–25% for 80M+ global listeners at ~45K events/sec on SNS/SQS/Lambda.
• Designed and built the OSI pipeline framework for Amazon Music's document ingestion, proposing per-cluster pipelines over per-index to reduce blast radius, lower OCU cost through better pipeline utilization, and simplify operations. Leading 3 engineers to build out the pipelines for live updates and backfills. Framework adopted by an adjacent team for their own use cases. Cut full-catalog backfill time ~60% (5 days → 2 days) and infrastructure cost ~33%.
• Diagnosed and fixed a search bug where less authoritative track variants were surfacing over the canonical version, by marking the authoritative variant at ingestion and adding a search-side filter on that field. A/B tested globally over 14 days, lifting precision by 7 percentage points for 50M+ North American listeners with no recall regression.
• Migrated search platform alarms from static to dynamic thresholds after an AZ outage revealed static thresholds had masked a real latency spike, closing a class of monitoring gaps.
• Built an internal admin tool for on-demand document indexing that replaced a manual runbook (SSH into a tools host, run ad-hoc scripts) with a self-serve interface used across engineering and content teams to push time-sensitive updates into search.

Software Engineer II

• Built per-PR CI infrastructure that provisioned APIC simulator fabrics on demand, replacing a manual, sprint-cadence test process.

Software Engineer

• Joined as the 4th engineer, working directly with clients to shape product requirements and building the core mortgage lending platform that grew the company from 4 to 20+ employees.
• Built an SMS-based mortgage system that shifted the entire loan workflow from desktop to phone, letting lenders process loan applications by text from anywhere.

Blockchain for Medical Records

Built a distributed system where processes compete across a network to mine and verify blocks (medical records) before being added to a ledger. Implemented from scratch using inter-process communication, cryptography, multithreading, and peer-to-peer networking.

IPC  ·  Cryptography  ·  Multithreading  ·  P2P Networking

Custom Memory Manager

Implemented a custom memory manager in C++ that replaces the standard library allocator, achieving a 3.5x speed increase through optimized allocation strategies and cache-aware data layout.

C++  ·  Memory Management  ·  Performance Optimization

Movie Recommendation System

Built a web application that recommends movies using collaborative filtering (user-based and item-based) and K-nearest neighbor, generating personalized recommendations based on a target user's ratings and their top-K similar users.

Collaborative Filtering  ·  KNN  ·  Flask  ·  Recommendation Systems

M.S. in Computer Science

B.S. in Computer Engineering  ·  Minor: Computer Science