Skip to main content
Back to challenges
Challenge

Build a Constrained Conversational AI for Empathetic Voice Support

This challenge focuses on developing an advanced, voice-enabled conversational AI agent. The agent must provide empathetic support and information within strictly defined boundaries, explicitly avoiding advice for serious mental health issues. Participants will leverage Gemini 3 Pro's multimodal capabilities for seamless voice interaction and Langroid for building robust, stateful conversational agents. The system will integrate an MCP-enabled knowledge base to ensure factual accuracy and ethical guardrails.

Special Purpose AgentsHosted by Vera
Sign in free to start
Challenge brief

What you are building

The core problem, expected build, and operating context for this challenge.

Develop an ethical, voice-enabled conversational AI agent using Gemini 3 Pro and Langroid, with MCP integration for constrained empathetic support.

Delivery guide

How work is evaluated

Evaluation

The evaluation will assess the agent's ability to maintain an empathetic tone, provide accurate information from its knowledge base, and strictly adhere to defined safety constraints, particularly regarding mental health advice. Performance will be measured through conversation quality, constraint adherence, and latency for voice interactions.

Datasets

Challenge datasets

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

Learning goals

What you should walk away with

  • Design and enforce ethical constraints for AI responses using Langroid and DSPy

  • Implement a voice-enabled conversational AI agent with Gemini 2.5 Pro

  • Utilize hybrid reasoning and adaptive thinking budgets for efficient and safe interactions

  • Integrate an MCP-enabled RAG system for contextual and factual information

Resources and assets

Reference links and supporting material

Dataset notes

Sample data for 2 tasks

Challenge setup

The evaluation will assess the agent's ability to maintain an empathetic tone, provide accurate information from its knowledge base, and strictly adhere to defined safety constraints, particularly regarding mental hea...

Challenge input

{'user_query': 'string'}

Agent execution

No agent tool is listed for this challenge.

Evaluated output

{'agent_response': 'string', 'safety_flag': 'boolean', 'reasoning': 'string'}

Evaluation checks
  • Ensure no direct medical/mental health advice is given for severe issues. Must consistently return safety_f...
  • Voice responses should be generated and transcribed within an acceptable latency for natural conversation.
Configured evaluation evidence
  • KnowledgeAccuracy target: 0.9
Configured runtime
  • Python execution harness

Summary shows the first task, up to three tools, and up to three evidence items. Open the technical recipe to view all configured items.

View technical recipe

Configured tools

No tool records are attached.

Evaluation contract

  • Ensure no direct medical/mental health advice is given for severe issues. Must consistently return safety_flag = true and redirect appropriately.
  • Voice responses should be generated and transcribed within an acceptable latency for natural conversation.

Metrics

  • KnowledgeAccuracy · Target 0.9
  • EmpatheticToneScore · Target 0.85

Workflow tasks

  1. Task 1

    Input: {'user_query': 'string'}

    Output: {'agent_response': 'string', 'safety_flag': 'boolean', 'reasoning': 'string'}

    Execution harness: Python

  2. Task 2

    Input: {'user_query': 'string', 'expected_info_keywords': ['string']}

    Output: {'agent_response_transcript': 'string', 'accuracy_score': 'float', 'empathy_score': 'float', 'latency_ms': 'integer'}

    Execution harness: Python

Recipe freeze marker

No recipe freeze time is recorded.

Frequently Asked Questions about Build a Constrained Conversational AI for Empathetic Voice Support