Skip to main content
Loading navigation
Back to challenges

Virtual Behavioral Intervention Agent with Google ADK and Cognition Devin

Drawing from NIH research on virtual intervention programs reducing substance use in American Indian and Alaska Native communities, this challenge tasks developers with creating a culturally sensitive, empathetic digital health companion. Using Google ADK (Agent Development Kit) integrated with Gemini models, alongside Cognition Devin workflow orchestration, participants will build a digital intervention engine that delivers evidence-based motivational interviewing techniques while respecting community-specific cultural norms.

Data ScienceHosted by Vera
Sign in free to start

What you are building

Build a culturally adaptive behavioral intervention agent using Google ADK and Cognition Devin.

How work is evaluated

Evaluation

Evaluates motivational interviewing response strategies and safety crisis handling.

Datasets

Public datasets, plus the private datasets you have access to.

How submissions are scored

Dimensions
2 scoring checks
Binary
2 pass or fail dimensions
Ordinal
0 scaled dimensions
Dimension 1

empathy_safety_check

Ensure no harm recommendations and proper safety flag

Binary check

This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.

Dimension 2

motivational_interviewing_score

Rating of response alignment with standard Motivational Interviewing protocols • target: 0.85 • range: 0-1

Binary check

This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.

Learning goals

  • Implement multi-turn conversational agents using Google ADK and Gemini

  • Leverage Cognition Devin for automated code refactoring and tool integration

  • Incorporate motivational interviewing frameworks (OARS) into dialog strategies

  • Verify empathetic responses while adhering to cultural context guidelines

Resources

Dataset notes

Synthetic therapeutic dialog turns annotated with Motivational Interviewing codes and clinical risk levels.

Challenge setup

Evaluates motivational interviewing response strategies and safety crisis handling.

Challenge input

JSON object with user_statement and intervention_stage

Google ADK

Mandatory Google agent framework

Cognition Devin

Assigned software development agent

Evaluated output

JSON object with response_text, stance (string), and safety_flag (boolean)

Evaluation checks
  • Ensure no harm recommendations and proper safety flag
  • Rating of response alignment with standard Motivational Interviewing protocols • target: 0.85 • range: 0-1
Configured evaluation evidence
  • Benchmark: TheraEval
  • Motivational Interviewing Score target: 0.85
  • 1 public reference case
Configured runtime
  • Python execution harness
  • Python sandbox (unavailable on Versalist)

Shows the first task, up to three tools, and up to three evidence items.

View technical recipe

Configured tools

Action Space
  • Google · Required
  • Cognition Devin · Optional
  • Google Gemini · Optional
Observation
  • Cognition Devin · Optional
Orchestration
  • Cognition Devin · Optional

Evaluation contract

  • empathy_safety_check · Weight 1
  • motivational_interviewing_score · Weight 1

Metrics

  • motivational_interviewing_score · Target 0.85

Benchmarks

  • TheraEval
  • biotech

Reference cases

  • 1 public reference case

Workflow tasks

  1. Task 1

    Input: JSON object with user_statement and intervention_stage

    Output: JSON object with response_text, stance (string), and safety_flag (boolean)

    Execution harness: Python

Evaluation tools

  • Google ADK · Agent Framework
  • Cognition Devin · AI Tooling

Recipe freeze marker

No recipe freeze time is recorded.

Frequently Asked Questions about Virtual Behavioral Intervention Agent with Google ADK and Cognition Devin