aws-samples/sagemaker-genai-hosting-examples
Apr 2026 – May 2026
2 Months active
Languages Used
MarkdownPython
Technical Skills
AWSAWS SageMakerData ScienceDeep LearningDockerMachine Learning
Manuwai Korber
PROFILE
Manuwai Korber
Over a two-month period, contributed to the aws-samples/sagemaker-genai-hosting-examples repository by developing and deploying advanced NLP and machine learning solutions using Python and AWS SageMaker. Delivered end-to-end workflows for deploying DeBERTa models with ONNX and Triton Inference Server, including a two-stage DeBERTa and XGBoost ensemble for scalable inference. Implemented a zero-shot email threat classification system leveraging TensorRT FP16 and ONNX optimizations to enable efficient, production-ready inference without Python in the hot path. Enhanced deployment scaffolding, documentation, and instance configuration, ensuring reproducible pipelines and providing clear guidance for real-world, production-oriented AI and data science deployments.
Overall Statistics
Feature vs Bugs
100%Features
Repository Contributions
10Total
Bugs
0
Commits
10
Features
4
Lines of code
3,064,725
Activity Months2
Your Network
46 peopleShared Repositories
46
Andrew SmithMember
Ayush SharmaMember
btinarwoMember
Christian-kamMember
Davide GallitelliMember
dlraghaMember
Dan-oMember
Eashan KaushikMember
EC2 Default UserMember
Work History
May 2026
1 Commits • 1 Features
May 1, 2026May 2026: Delivered a zero-shot email threat classification system (DeBERTa + XGBoost ensemble) with end-to-end, production-ready inference optimized for performance and deployment. Implemented an NLI cross-encoder scoring against a 50-category threat taxonomy to classify emails as malicious or benign, with a lightweight XGBoost head for final decision. Achieved significant runtime efficiency by leveraging TensorRT FP16 and ONNX graph optimizations, enabling end-to-end inference on Triton's onnxruntime backend with no Python in the hot path. Added comprehensive production deployment notes and ahead-of-time planning guidance. Reworked the original example to align with email threat detection, improving clarity and applicability for real-world deployments.
1 Commits • 1 Features
May 1, 2026May 2026: Delivered a zero-shot email threat classification system (DeBERTa + XGBoost ensemble) with end-to-end, production-ready inference optimized for performance and deployment. Implemented an NLI cross-encoder scoring against a 50-category threat taxonomy to classify emails as malicious or benign, with a lightweight XGBoost head for final decision. Achieved significant runtime efficiency by leveraging TensorRT FP16 and ONNX graph optimizations, enabling end-to-end inference on Triton's onnxruntime backend with no Python in the hot path. Added comprehensive production deployment notes and ahead-of-time planning guidance. Reworked the original example to align with email threat detection, improving clarity and applicability for real-world deployments.
May 2026
April 2026
9 Commits • 3 Features
Apr 1, 2026In April 2026, delivered end-to-end deployment workflows for DeBERTa on SageMaker using ONNX/Triton, introduced a two-stage DeBERTa + XGBoost ensemble, and deployed Devstral-Small-2-24B-Instruct-2512 via SageMaker vLLM DLC. Implemented SDK v3 migrations, performance-tuned configurations, and robust scaffolding with reusable scripts and updated documentation. This work provides reproducible, scalable inference pipelines with GPU/CPU options and real-time chat capabilities, accelerating time-to-value for complex AI deployments.
April 2026
9 Commits • 3 Features
Apr 1, 2026In April 2026, delivered end-to-end deployment workflows for DeBERTa on SageMaker using ONNX/Triton, introduced a two-stage DeBERTa + XGBoost ensemble, and deployed Devstral-Small-2-24B-Instruct-2512 via SageMaker vLLM DLC. Implemented SDK v3 migrations, performance-tuned configurations, and robust scaffolding with reusable scripts and updated documentation. This work provides reproducible, scalable inference pipelines with GPU/CPU options and real-time chat capabilities, accelerating time-to-value for complex AI deployments.
Activity
Loading activity data...
Quality Metrics
Correctness100.0%
Maintainability88.2%
Architecture100.0%
Performance92.0%
AI Usage50.0%
Skills & Technologies
Programming Languages
MarkdownPython
Technical Skills
AWSAWS SageMakerData ScienceDeep LearningDockerMachine LearningModel DeploymentNLPNatural Language ProcessingSageMakerTriton Inference Serverdata engineeringdocumentationmachine learningmodel deployment
Repositories Contributed To
Overview of all repositories you've contributed to across your timeline