Wilson K. Choi

Joined the MLOps team to build out the model packaging and deployment pipeline that takes trained artifacts from Data Science and packages, registers, and deploys them to AWS SageMaker for real-time and batch inference.

• Pipeline Contribution: Contributed to the Jenkins + Python pipeline that validates trained scikit-learn artifacts, repackages them via a SageMaker training job, registers them in Model Registry, and deploys to real-time endpoints and batch transform across regions.
• Load Testing: Integrated SageMaker Inference Recommender into the pipeline for automated load testing, producing capacity and latency reports (stored to S3) to right-size endpoint configuration before production.
• Observability: Designed and implemented end-to-end model-serving observability, streaming SageMaker and endpoint metrics through CloudWatch to on-prem ELK and Grafana dashboards to monitor model latency and endpoint health.
• Nightly Builds: Introduced "Nightly Builds" running the full packaging and deployment flow against a synthetic test model, catching pipeline and artifact regressions before real model releases, with automated resource cleanup.
• Concurrency Safety: Built a Python concurrency safety test harness to catch subtle multithreading bugs in inference code handed off from Data Science, preventing race conditions in production model serving. Later generalized into an open-source tool (see Projects).

Embedded across multiple teams as a technical resource to deliver new products, address system performance issues, and establish operational standards.

• Push Verify: Built a gRPC-based push notification verification service (Python, RabbitMQ, Firebase FCM) with multi-region deployment, operational runbooks, and Grafana/ELK monitoring.
• Score Architecture: Architected, designed, and implemented observability (ELK APM) for the Score fraud-detection serving layer, and authored the Score 2.0 architecture proposal and handover documentation that gave the inheriting team a clear implementation path.
• Email Cleansing: Developed a Python email validation and normalization library using Parsing Expression Grammar (PEG), replacing a regex approach that grew unmaintainable as provider-specific rules (Google, Yahoo, Outlook) multiplied. Supports RFC 5322 compliance.

Architected and built the "Voice CPaaS" platform ("Vox"), transforming a legacy VoIP stack into the company's primary revenue engine.

• System Rebuild: Migrated an Asterisk stack to FreeSwitch, Kamailio, and FastAPI running on-prem and AWS, enabling features including IVR, Text-to-Speech, Conference Bridging, and Automated Speech Recognition (ASR) for enterprise customers including Microsoft (Azure).
• Scaling & Availability: Designed a VoIP architecture with a Kamailio SIP proxy, scaling platform capacity from 50 to 300 new calls per second (CPS) while maintaining 99.99% availability across US, EU, and Asia data centers.
• Hybrid Cloud & Call Recording: Engineered a hybrid architecture integrating on-premise VoIP system with AWS (Polly, S3, CloudFront, SQS, DynamoDB) and GCP (Speech-to-Text). Designed end-to-end Call Recording & Transcription pipeline (FreeSwitch → S3 → Transcription → customer-facing API).

Led a squad of 4 engineers building high-volume fraud detection APIs. Served as both Technical Lead and People Manager.

• Performance Engineering: Reduced Risk Scoring API latency from 300ms to 50ms (-83%) by optimizing data retrieval logic and implementing a multi-layer caching strategy.
• Data Pipeline: Managed the ETL pipeline processing millions of daily transactions, serving as the backbone for the company's "Phone Number Intelligence" product.
• Leadership: Mentored team members, conducted performance reviews, and managed product architecture planning. Delivered the "Batch Scoring Processor," which unlocked a new enterprise customer segment.

• Internal Tools: Built centralized dashboards (Django) to manage API features, BI reporting, customer configurations, and billing/invoicing (NetSuite integration).
• Data Warehouse: Collaborated with DBAs to design and implement OLTP/OLAP systems and ETL processes, improving reporting capabilities for the sales and business intelligence teams.

Early engineering hire (#50); company grew from startup to 700+ employee global enterprise during tenure.

• High-Throughput Messaging: Implemented asynchronous queuing systems using Python Celery and Redis, resolving critical bottlenecks in transaction capacity.
• Gateway Integration: Integrated backend systems with Kannel (Open Source WAP/SMS gateway) to expand carrier connectivity.
• Distributed Systems: Refactored the monolithic backend to support distributed compute requirements, migrating from a single data center to a multi-region (3 DC) architecture.

• Protocol & Messaging System: Built a custom SMPP (Short Message Peer-to-Peer) client in pure Python from scratch.
• Billing Systems: Built an SMS billing system with messaging queue and async Python worker to communicate with the SMS center over SMPP.
• Mobile WAP Development: Worked on in-house web framework, internal tools, third party service integration (REST, SOAP, raw socket).