Continual, Adaptable, Dynamic AI  systems


Navigating streams

Regardless of industry, A.I systems must navigate the stream of data your company feeds it to generate insights for your business or enhance the customer product. However, many A.I systems are brittle in nature and require large-scale periodic revamping/adjustment. In other cases they can only navigate very specific "stream conditions" and break down when the conditions change even slightly. Below we overview the conditions streams commonly have and how A.I. models should navigate them.

Streams flow continuously

Streams are continually moving, changing, and dynamic. New rocks, sediment, and intense rain can change their flow.  Even without flash floods, streams slowly erode the river and cut new paths. Models should similary, adapt to changes, continually improve their performance and grow in the number and type of tasks they can accomplish while still gracefully maneuvering down the old waterways when needed. 

Small, uncharted streams

While there are massive rivers like the Nile, Amazon, Yangtze,  Colorado,  and the Congo, the number of small brooks is much greater. Similarly, the majority of A.I. algorithms presume "big and massive datasets"  when the majority of industries do not have large datasets.  "Real A.I." systems should learn to rapidly adapt to these small and sometimes unchartered micro-sets of data.

Streams are turbulent

Streams often have boulders, rock, and drops that cause turbulence.  Data ingested by A.I. often has missing values, inconsistent intervals, and misspelled words/names. We understand this problem. A.I. models must gracefully navigate through the turbulent and inconsistent data.

Streams have unique places

Streams have certain crucial areas that characterize them. A stream can meander for hundreds of miles, and have one significant waterfall that defines it. Similarly, data has specific important areas that an A.I model must learn to attend to. To this point many A.I. models cannot utilize the natural structure of the data ingested. Thus, models require more data to learn patterns (and longer training times). A.I. algorithms need the ability to effectively learn from existing structured data structures so they maneuver precisely over the edge and land softly at the base of the waterfall of your information.

Streams can have multiple different sources

Streams can get their water from lakes, ponds, melting snow/glaciers, rainstorms, smaller tributaries or a mix of the above. In the same way your data can flow from a number of different sources in a number of different formats. Whether your data is historical data in a SQL database, realtime data continuously transmitted from around the world or unstructured data in a data lake A.I systems must navigate the information delta effectively. Moreover, A.I should also effectively utilize  the different modalities of sources such as images, text, audio, and signals, all within the same model.

Streams can move rapidly or barely at all

A.I. models should be able to handle either processing a few requests an hour or hundreds of requests per second.  They should be able to rapidly scale up or down based on your need.  There is no need to have an A.I. model running on multiple GPUs if you are only processing a few predictions per hour as this wastes valuable money. In contrast, if models need micro-second latency they should respond that quickly.



We are located in the scenic small town of Orono Maine. Orono is an outpost on the edge of a the great Maine outdoors. Despite being small, Orono is home to many restaurants and a large research university (whom we actively partner with).  However, we are open to remote work as well. We currently do not have the resources for sponsoring visas. Therefore we can only accept applications from U.S. citizens and green card holders. If any of the jobs below interest you please send your resume to with the subject the position that you are applying to.

Software Engineer

Onsite or Remote

  • Core Skills

    • Solid Knowledge of Data Structures/Algorithms

    • 2+ years of OOP experience 

    • Docker/Kubernetes experience and understanding of micro-services

    • Experience buildings rest APIs 

    • Experience with big data processing frameworks Flink, Spark, or Hadoop.

    • Solid SQL and database design skills

    • Experience with AWS/Google Cloud/Azure

  • Other skills

    • Python and/or Java ​experience 

    • Django/Flask experience 

    • Experience using Tensorflow Serving or ONNX to deploy models

    • Knowledge of Kafka or similar streaming technologies

A.I. Researcher 

Onsite or Remote

  • Core Skills

    • Excellent Python​

    • Experience with Tensorflow/PyTorch/Keras or other DL framework 

    • Docker/Kubernetes experience 

    • Publications at NIPS, ICML, ACL, or other top A.I. conference.

    • Basic knowledge of non deep-algorithms/techniques like XGBoost, Random Forests, ensembling 

  • Other skills

    • We are particularly interested in researchers with research in one of the following areas

      • Meta-learning ​

      • Continual learning 

      • Transfer learning/Domain adaptation 

      • A.I. interpretability

      • Back-propagation through structure 

Data Visualization Engineer

Onsite or Remote

  • Core Skills

    • Solid Knowledge of Data Structures/Algorithms​ including geo-data structures

    • 2+ years of OOP experience 

    • Docker/Kubernetes experience and understanding of micro-services

    • Experience buildings rest APIs 

    • Experience with Bokeh/D3.js or similiar visualization framework.

    • Experience with Python and/or JavaScript

  • Other skills

    •  Publications at IEEE DataVis or similar conference

    • Django/Flask experience 

    • Experience visualizing geo temporal data

    • Knowledge of Kafka or similar streaming technologies


More info

We are currently in stealth mode and are purposely "keeping quit."  Check back on our site periodically as we will display more information about our research as well as our launch. If you want to chat with us in private you are welcome to contact us using one of the methods below.

Skype Username
Also, drop by us at PyData Orono Meetups and Events.

4 Lane Orono, ME 04473

  • White LinkedIn Icon
  • White Facebook Icon
  • White Twitter Icon
  • White Google+ Icon

© 2023 by AI Stream. Proudly created with