Peter Rigby: Experiences designing and deploying AI-assisted coding tools at Meta

mercredi 11 décembre 2024 - Séminaires

Heure et date : le 11 décembre 2024 à 12h45
Lieu : PK-4610 et Zoom
Conférencier : Peter Rigby (Concordia)
Titre : Experiences designing and deploying AI-assisted coding tools at Meta

Veuillez prendre note que ce séminaire ne sera pas enregistré.

—

Date and time: December 11, 2024 at 12:45 PM
Location: PK-4610 et Zoom
Speaker: Peter Rigby (Concordia)
Title: Experiences designing and deploying AI-assisted coding tools at Meta

Please note that this seminar will not be recorded.

—

Abstract:
I will cover various challenges encountered during the development and deployment of AI-assisted code authoring solutions at industry scale. Covered topics include:

An overview of the design and implementation of CodeCompose, Meta’s internal copilot

  1. How Meta achieved multi-line AI-assisted code generation
  2. The challenges and solutions behind SQLCompose, a copilot specifically tuned for SQL
  3. Fine-tuning AI models for industry-specific applications
  4. The deployment process of AI-assisted tools
  5. Evaluation methods, including mixed-methods approaches for large-scale AI code generation systems
  6. Using AI models to predict commit risk

The papers covered in this talk are

  1. AI-assisted Code Authoring at Scale: Fine-tuning, deploying, and mixed methods evaluation (CodeCompose)
  2. Multi-line ai-assisted code authoring
  3. AI-Assisted SQL Authoring at Industry Scale (SQLCompose)
  4. Moving Faster and Reducing Risk: Using LLMs in Release Deployment

Bio:

Peter C. Rigby is a Software Engineering researcher at Meta and an associate professor in Software Engineering at Concordia University in Montreal. His overarching research interest is in understanding how developers collaborate to produce successful software systems. His research program is driven by a desire to determine empirically the factors that lead to the development of successful software and to adapt, apply, and validate these techniques in different settings. Empirical Software Engineering involves mining large data sets to provide an empirical basis for software engineering practices. Software Analytics is then used to provide statistical predictions of, for example, the areas of the system that would benefit from increased developer attention. Grounded, empirical findings are necessary to advance software development as an engineering discipline. He is currently focusing on the following research areas: software testing, developer turnover and knowledge loss, and code review. He has extensive industry collaboration with Ericsson and Meta.

https://www.linkedin.com/in/peter-rigby-27532a7b
https://users.encs.concordia.ca/~pcr/