Jason Yang
Computer Vision
Management: Under a year
Python
TypeScript
SQL
Node JS
C++
JavaScript
C#
C
MATLAB
Data Pipeline & Feature Engineering: 5/10
MLOps & Deployment: 5/10
Computational Efficiency: 5/10
Active 5 days ago
Download CVCV
Overview
Technical skills
Projects
Timeline
Roles
Overview
A senior ML practitioner specializing in production image inference and computer vision pipelines. The strongest proven skill is building production inference pipelines with batching, Redis-backed queues, ONNX and PyTorch backends plus Prometheus observability as shown in app/worker.py and app/onnx_inference.py. There is limited evidence of rigorous statistical evaluation, systematic model selection, or enterprise MLOps automation such as data-versioning, CI pipelines, or formal cross-validation workflows.
Technical skills
Languages
9
Python
TypeScript
SQL
Node JS
C++
JavaScript
C#
C
MATLAB
Python
5
SQLAlchemy
Flask
HTTPX
Asyncio
FastAPI
AI/ML
12
NumPy
OpenCV
Pandas
Scikit-learn
ONNX Runtime
Pillow
Streamlit
CUDA
Hugging Face
Torchvision
CLIP
PyTorch
DevOps
4
API Gateway
AWS Lambda
AWS
Prometheus
Other
19
React.js
PostgreSQL
Redis
Axios
AI/ML
Deep Learning
GitHub
Grafana
Docker
Docker Compose
Git
Rest API
ESLint
ONNX
Machine Learning
Unix
Babel
Computer Vision
HPC
Projects
Builds an experimental soybean segmentation pipeline for field images using unsupervised clustering and color-space features to identify vegetation regions. Integrates a pretrained BioCLIP vision model to score plant regions and refine segmentation masks in scenes with weeds and grass. Performs perspective correction and exports per-image segmentation masks, overlays, and CSV scores for review.
Creates a containerized AI inference service for image classification with an API backend and a React-based interface. Implements asynchronous job queues with dynamic GPU batching, using PyTorch and ONNX Runtime backends with model versioning. Adds performance benchmarking and operational monitoring using Prometheus and Grafana.
Timeline
Computer Vision & Machine Learning Research Intern
·
Junior
Weed Ecology and Management Lab
·
Internship
Develops machine learning models to analyze RGB and stereo depth camera images for automated plant trait extraction. Works with large-scale greenhouse and field datasets collected via the PlantMap3D camera system. Trains and optimizes computer vision models using HPC resources and collaborates with PhD students and lab mentors to advance the image analysis pipeline.
Machine Learning
Telemetry Project Team Lead
·
Lead
Buckeye Solar Racing
·
Full-Time
Leads development of a solar car telemetry system for real-time vehicle performance monitoring. Builds a full-stack telemetry platform with React for visualization and Flask with SQL for data handling. Designs and deploys AWS cloud infrastructure using Lambda and API Gateway to support scalable storage, processing, and access to live telemetry data.
React.js
Flask
SQL
AWS Lambda
API Gateway
Teaching Assistant (CSE 2221: Software I)
·
Middle
The Ohio State University
·
Full-Time
Supports students in an introductory software engineering course focused on Java development. Supervises lab work and helps guide students through troubleshooting and problem-solving tasks. Provides individualized assistance for Eclipse-based development assignments and classroom exercises.
Senior AI/ML Engineer
Confidence: High ML Engineer
ML engineer focusing on production inference and computer vision pipelines at a senior level with strong practical system-building skills. The strongest proven skill is building end-to-end inference and deployment pipelines, evidenced by the FastAPI + Redis worker + ONNX/PyTorch backends and benchmarking/metrics code (app/worker.py, app/onnx_inference.py, benchmarks/benchmark.py). Public code shows careful CV preprocessing and heuristics, but lacks custom large-scale training pipelines, experiment tracking instrumentation, and broader deployment artifacts (some README-described components such as a React dashboard or Docker Compose are not present in code).
Model Architecture & Training
3/10
How well models are designed and trained
Model architecture and training focus is on model consumption and conversion rather than custom training; clear use of pretrained vision models, ONNX export/serving and basic classical ML pipelines.
Evidence
Production-AI-inference-platform/app/inference.py: ImageClassifier using torchvision resnet18 and ResNet18_Weights
Production-AI-inference-platform/app/onnx_inference.py: ONNX runtime session loading and inference flow
Production-AI-inference-platform/scripts/export_onnx.py: (present in repo dependencies and scripts) exporting PyTorch model to ONNX
Data Pipeline & Feature Engineering
5/10
How data is prepared for models
Nontrivial data preprocessing and feature engineering for computer vision tasks, with explicit color-space transforms, shape filtering and clustering heuristics.
Evidence
Bioclip_segmentation/unsupervised_clustering/soybean_clustering.py: features(), compute_exg and filter_leaf_shapes
Bioclip_segmentation/unsupervised_clustering/bioclip_segment.py: make_vegetation_mask(), split_vegetation_regions(), refine_accepted_region()
Experimentation & Evaluation
4/10
How results are measured and tested
Has benchmarking and basic test artifacts for latency and queue behavior, but lacks structured experiment tracking (W&B/MLflow) or extensive reproducible experiment harnesses.
Evidence
Production-AI-inference-platform/benchmarks/benchmark.py: async latency benchmarking and percentile reporting
Production-AI-inference-platform/benchmarks/queue_benchmark.py: end-to-end async queue benchmarking
Production-AI-inference-platform/tests/test_api.py: basic FastAPI endpoint tests
MLOps & Deployment
5/10
How models are shipped to production
Clear production-oriented serving architecture with FastAPI endpoints, Redis-backed async queue, worker process, ONNX and PyTorch backends, and Prometheus metrics integration.
Evidence
Production-AI-inference-platform/app/worker.py: worker loop, dynamic batching, Redis queue interactions and metrics
Production-AI-inference-platform/app/onnx_inference.py: ONNX provider selection and runtime configuration
Production-AI-inference-platform/app/api_prometheus.py: Prometheus metrics usage and instrumentation
Computational Efficiency
5/10
How efficiently computing resources are used
Practical efficiency work is visible: dynamic request batching, CUDA synchronization, numpy/onnx numeric handling, and benchmarking tooling for throughput/latency measurement.
Evidence
Production-AI-inference-platform/app/worker.py: collect_batch batching logic and use of MAX_BATCH_SIZE/BATCH_WAIT_MS
Production-AI-inference-platform/app/inference.py: batch preprocessing and torch.cuda.synchronize usage
Production-AI-inference-platform/app/onnx_inference.py: numpy stacking and numerically stable softmax for ONNX outputs
Research Depth & Innovation
4/10
Depth of research and new ideas
Demonstrates domain-aware algorithmic engineering (segmentation heuristics, hybrid BioCLIP+morphology pipeline) but not novel research contributions or reproduced SOTA experiments.
Evidence
Bioclip_segmentation/unsupervised_clustering/bioclip_segment.py: hybrid pipeline integrating BioCLIP region classification with morphology heuristics
Bioclip_segmentation/unsupervised_clustering/soybean_clustering.py: clustering + shape-based refinement and careful heuristics
Verified artifacts
Expertise
Computer Vision & Image Analysis· Senior
MLOps & Model Lifecycle· Senior
AI Infrastructure & Optimization· Senior
Smart Agriculture & AgTech· Middle
Industries
Artificial Intelligence· Senior
Farming & Agriculture· Senior
Health Care· Middle
Technologies
SQL· since 2025 · Junior
C++
MATLAB
Computer Vision
ONNX
Git
PyTorch
AWS
Docker
Grafana
HTTPX
CLIP
Torchvision
AWS Lambda· since 2025
GitHub
API Gateway· since 2025
HPC
Unix
Machine Learning· since 2026
Recommendations
- Implement and own inference-serving components: build scalable FastAPI + worker deployments with Prometheus metrics and ONNX/PyTorch model backends.
- Design and maintain computer vision data pipelines and segmentation workflows for agriculture or remote-sensing projects where careful preprocessing and heuristics are required.
- Lead benchmarking and performance tuning efforts: optimize batching, provider selection, and GPU/ONNX runtime choices using the existing benchmark suite.
- Extend MLOps with experiment tracking and CI/CD for model lifecycle (W&B or MLflow, reproducible export pipeline and deployable Docker/Kubernetes manifests).
Repositories
The developer's experience in this domain has been verified based on AI analysis of the following repositories:
Senior Data Scientist
Confidence: Medium ML Practitioner
A senior ML practitioner specializing in production image inference and computer vision pipelines. The strongest proven skill is building production inference pipelines with batching, Redis-backed queues, ONNX and PyTorch backends plus Prometheus observability as shown in app/worker.py and app/onnx_inference.py. There is limited evidence of rigorous statistical evaluation, systematic model selection, or enterprise MLOps automation such as data-versioning, CI pipelines, or formal cross-validation workflows.
Statistical Rigor
2/10
Correct use of statistics
Statistical reasoning and rigorous evaluation are minimal; models are trained and basic metrics reported without cross-validation, uncertainty quantification, or formal hypothesis testing.
Evidence
BDAA_DIabetes_Project/RandomForestReg.py: single train/test split and metric prints without CV or uncertainty estimates
BDAA_DIabetes_Project/jerrymain.py: basic accuracy and classification_report usage with no cross-validation or calibration
Data Wrangling & Cleaning
5/10
Preparing and cleaning data
Practical data wrangling and robust image preprocessing are well executed in the CV pipelines, while tabular-data cleaning in the diabetes project is basic and ad-hoc.
Evidence
Bioclip_segmentation/unsupervised_clustering/soybean_clustering.py: features, rectify, and filter_leaf_shapes implement extensive image cleaning and shape filtering
app/inference.py: predict and predict_batch include input validation and explicit error handling for invalid images
Exploratory Analysis & Visualization
6/10
Exploring and visualizing data
Exploratory analysis and visualizations are purposeful and tied to decision points in the CV work; dashboards and plotting code show attention to interpretable outputs rather than decorative charts.
Evidence
Bioclip_segmentation/unsupervised_clustering/soybean_clustering.py: save_panels and save_shape_comparison produce reviewable visual summaries and per-cluster stats
BDAA_DIabetes_Project/jerrydashboard.py: Streamlit dashboard with confusion matrix, correlation heatmap, and feature importance visualizations
Predictive Modeling
5/10
Building models that predict
Predictive modeling and deployment are solidly covered from prototype training to production inference and ONNX runtime paths; evaluation practices and model-selection workflows are present but not exhaustive.
Evidence
app/inference.py: ImageClassifier.predict and predict_batch implement batching, timing, and top-k result formatting
app/onnx_inference.py: ONNXImageClassifier implements provider selection, session setup, and numerically-stable softmax for batch inference
BDAA_DIabetes_Project/jerrymain.py: end-to-end training pipeline and model persistence with a preprocessing pipeline and RandomForestClassifier
Business Insight & Impact
2/10
Turning analysis into business value
Operational and performance metrics are captured (latency, throughput) but there is little explicit business-metric framing, cost-of-error analysis, or product-oriented decision rules.
Evidence
benchmarks/benchmark.py: measures latency percentiles and throughput for the serving API
app/worker.py: emits Prometheus metrics for batch and inference durations but no business-SLA to cost mapping in code
Reproducibility & Notebook Hygiene
4/10
Clean, repeatable analysis
Reproducibility and hygiene are present at a basic level (requirements, tests, deterministic RNG in places), but pinned environments, data versioning, and CI-level automation are not evidenced.
Evidence
requirements.txt: lists runtime dependencies
tests/test_api.py: basic API tests using fastapi.testclient
Bioclip_segmentation/unsupervised_clustering/soybean_clustering.py: uses a fixed RNG seed and joblib.dump for model persistence
Expertise
Data Science· Senior
Industries
Artificial Intelligence· Senior
Farming & Agriculture· Senior
Health Care· Middle
Technologies
AI/ML
Deep Learning
Python· since 2025 · Senior
PostgreSQL
Redis
OpenCV
SQLAlchemy
Scikit-learn
Pandas
NumPy
Asyncio
Streamlit
Pillow
CUDA
Hugging Face
ONNX Runtime
Recommendations
- Lead development of production image inference services that require batching, GPU/ONNX support, and operational metrics
- Build and harden model deployment and monitoring pipelines that integrate Redis job queues, Prometheus metrics, and ONNX/PyTorch backends
- Develop reproducible CV research-to-production workflows including CI, data versioning, and systematic evaluation (cross-validation and uncertainty estimates)
- Implement benchmarking and load-testing tooling for inference platforms and optimize batching and latency under realistic workloads
Repositories
The developer's experience in this domain has been verified based on AI analysis of the following repositories:
Middle Frontend Developer
Confidence: Medium UI Engineer
A frontend UI engineer at an early-middle level specializing in building polished responsive interfaces and mobile app UI. The strongest proven skill is crafting well-structured, responsive and accessible UI with theming shown in index.html, styles.css, and useThemeColor.ts. There is limited public evidence of advanced backend scalability design, measured performance work, server-state management, or comprehensive automated tests.
UI Component Architecture
4/10
How interface parts are built
Practical component and theming work across web and React Native with clear folder patterns and small reusable hooks, but no evidence of a bespoke component library, design tokens system beyond Colors, or advanced component composition at scale.
Evidence
dogbear24/dogbear24.github.io/index.html: structured semantic sections and component-like blocks (feature-project, secondary-project)
dogbear24/dogbear24.github.io/styles.css: large custom stylesheet organizing UI patterns and responsive layout
dogbear24/MakeOhio-2025-2/hooks/useThemeColor.ts: theme hook applying Colors across components
Responsive & Cross-browser
5/10
Works on all screens and browsers
Responsive-first CSS with multiple breakpoints, grid layouts and reasonable mobile fallbacks; not much evidence of RTL or advanced feature detection but clear responsive strategies in styles.
Performance Optimization
3/10
Speed of the interface
Small performance-aware choices (passive scroll listener, hydration-aware color-scheme hook) but no measured optimizations, bundle analysis, or advanced code-splitting; overall basic attention to runtime cost.
Accessibility & Semantics
5/10
Usable for everyone
Good use of semantic elements and ARIA hints for navigation and areas; keyboard/focus management on custom widgets is not visible and there is no CI a11y tooling evident.
State Management & Data Flow
2/10
Managing data in the app
Basic client-side data fetching and small hooks for theme; no evidence of robust server-state management, request cancellation, optimistic updates, cache invalidation, or state machines.
UX & Visual Polish
5/10
Look and feel quality
High visual polish and considered UX details across the portfolio site (hover states, info hierarchy, responsive visuals); UX engineering is strong for static and prototype UIs but lacks evidence of advanced interaction states like undo flows or skeleton loading components.
Expertise
React· Middle
Frontend Architecture & Build Tools· Middle
Industries
Artificial Intelligence· Middle
Software· Middle
Energy & Utilities· Junior
Technologies
JavaScript
TypeScript· since 2025 · Middle
Node JS· Middle
Docker Compose
Babel
Prometheus
React.js· since 2025
Axios
ESLint
Recommendations
- Own UI and theming tasks for responsive web sites and React Native screens, including design-to-code handoffs and component improvements.
- Work on build-tooling and developer-experience improvements such as TypeScript typing coverage, CI linting, and Babel/packager optimizations.
- Implement client-side state and data-flow patterns (React Query or similar) for robust caching, request cancellation, and optimistic updates in mobile apps.
- Develop small infrastructure automation scripts and dev-ops helpers (Node scripts, Docker Compose) for local dev and reproducible environments.
Repositories
The developer's experience in this domain has been verified based on AI analysis of the following repositories: