Statement of Work
This Statement of Work (SOW) describes Anote’s plan to extend its MLOps platform with robust computer vision and object detection capabilities, including model inference SDK, benchmarking research, documentation, and curated datasets. Anote will develop, evaluate, and deploy computer vision models for object detection, integrate them into a PyPI-distributed SDK, publish a benchmarking study, update documentation, and curate training/testing datasets aligned to a predefined ontology.
Objectives
- Deliver a production-ready inference SDK
- Benchmark object detection model performance
- Publish research on model comparisons
- Provide clear usage documentation
- Curate and maintain labeled image datasets
- Ensure scalability and extensibility
Milestones
| # | Milestone |
|---|---|
| 1 | Predict categories on test datasets from within the predict-table view using standard model providers. |
| 2 | Train models on domain training data via supervised and unsupervised fine-tuning. |
| 3 | Evaluate model performance to produce concrete metrics on held-out test data. |
| 4 | Incorporate subject matter expertise through the Data Labeler interface. |
| 5 | Implement RLHF fine-tuning, supervised fine-tuning, and any necessary pre-training to boost accuracy. |
| 6 | Integrate each model into the Software Development Kit (SDK) as an API endpoint. |
| 7 | Route the best-performing model into the chatbot for version testing in-context. |
| 8 | Scale the solution to adapt as training data volumes grow and business requirements evolve. |