1 of 9

Streamlining the OrcaAL (Orca Active Learning) tool

2 of 9

About myself

  • 3rd year Computer Engineering student from Singapore – currently at UC Berkeley on university exchange
  • Started my journey predominantly in data science and machine learning
  • Spent the last year or so improving my skills as a software engineer 🡪 GSoC’22!
  • Passionate about exploring ways to utilize technology to conserve wildlife

​

3 of 9

Background behind OrcaAL

  • GSoC’20 project by Diego Rodríguez and Kunal Mehta

​

Link to the tool: https://orcasound.github.io/orcaal/

4 of 9

Areas for improvement to be made to OrcaAL

  • Difficult to onboard new developers
    • Tests existed, but weren’t enforced
    • Many components of the tool to run separately 🡪 many languages and frameworks
  • Setup and code was very much tied to AWS S3, making it difficult for potential downstream applications building on top off OrcaAL

​

​

5 of 9

What I’ve done/been doing

  • Dockerized all services and tests
  • Image size from 7.52GB --> 2.2GB

6 of 9

What I’ve done/been working on

  • CI/CD with Github Actions

7 of 9

Further/Future Work

  • Improving test coverage (based on a paper titled CheckList) – behavioural and regression tests to measure model drift, etc.
  • Using OrcaAL’s infrastructure as code (IAC) to spin-off other active learning projects (Terraform?)

​

8 of 9

Any questions?

9 of 9

References

​