DEVELOP & DEPLOY FLOW PHOTO EXPLORER MODEL

The USGS Eastern Ecological Research Center, Conte Fish Research Laboratory in Turners Falls, MA, seeks assistance to develop and deploy a machine learning model to predict streamflow from timelapse images of streams as part of the USGS Flow Photo Explorer (FPE) project. This model, which is being developed by USGS in collaboration with Microsoft AI For Good using the PyTorch machine learning framework, is being trained using an existing database of images and human annotations of pairwise image rankings.

The FPE project is currently hosted using Amazon Web Services (AWS) through the USGS Cloud Hosting Solutions (CHS) platform.

Services are required for two tasks as outlined in Attachment 1, Performance Work Statement, 1) assist in the development of an automated machine learning pipeline that will facilitate model training and inference on a station-by-station basis, and 2) to develop a real-time inference endpoint of a trained model specific to a single station that can be used to predict streamflow from new images transmitted to FPE through cellular or satellite connections.

For both tasks, the contractor will focus solely on developing the necessary components within AWS SageMaker, namely a SageMaker Model Building Pipeline and a SageMaker Real-time Inference Endpoint (or possibly using alternative AWS computing service such as EC2 or Batch). The USGS will be responsible for developing the remaining project infrastructure to handle data ingest, storage, and access control, much of which already exists.

Information Technology Security (FISMA) requirements and Section 508 compliance are outlined in Attachments 2 through 5. This will be a firm fixed-price contract with value less than $30,000. Quote submissions are due on or before 04/19/2024. The anticipated award date approximately 5/01/2024. The acquisition is a total small business set-aside. The NAICS code is 541512 and size standard is $34 million. Vendors must be registered at https://www.sam.gov/ to be awarded the contract. All responsible sources may submit a quotation which shall be considered by the government. All correspondence and quotes MUST be emailed to sruggles@usgs.gov.