Adriana Voss

Profile

Senior ML Engineer with 8 years building and shipping production-grade models across NLP, recommendation systems, and computer vision at companies ranging from Series B startups to Fortune 500 firms. Deep expertise in transformer fine-tuning, MLOps pipeline design, and cross-functional collaboration with product and data engineering teams. Reduced model inference latency by 61% and lifted revenue-attributed recommendation CTR by 22% in the last two roles.

Work Experience

Senior Machine Learning EngineerMar 2021 - Present

Luminary AI, Austin, TX

Architected a real-time product recommendation engine using two-tower neural networks on PyTorch, increasing click-through rate by 22% and driving $4.1M in incremental annual revenue.
Reduced transformer inference latency from 340ms to 133ms by implementing dynamic quantization and TensorRT optimization, enabling a customer-facing feature previously blocked by SLA constraints.
Built an end-to-end MLOps platform on Kubernetes and MLflow that cut model deployment cycle time from 3 weeks to 4 days, supporting 14 models across 6 product teams.

Machine Learning EngineerJun 2018 - Feb 2021

Helix Data Systems, Denver, CO

Developed a document classification pipeline using fine-tuned BERT that achieved 94.3% F1 on a 22-class legal taxonomy, replacing a rules-based system that topped out at 71% accuracy.
Designed and maintained feature engineering workflows in Apache Spark processing 900GB of daily event data, reducing pipeline runtime by 38% through partition pruning and broadcast join tuning.
Mentored 3 junior engineers through weekly code reviews and pair-programming sessions, two of whom were promoted within 18 months.

Machine Learning Engineer IAug 2016 - May 2018

Stackframe Analytics, Boulder, CO

Built a churn prediction model using gradient-boosted trees (XGBoost) that identified at-risk accounts with 87% precision, enabling the customer success team to retain $1.2M in ARR in Q3 2017.
Automated model retraining and validation using Airflow DAGs, reducing manual intervention from 6 hours per week to under 30 minutes.
Collaborated with data engineering to migrate training pipelines from on-premise Hadoop to AWS SageMaker, cutting infrastructure costs by 44%.

Profile

Work Experience

Education

Skills

Links