Skip to main content
Back to challenges
Challenge

Build OpenAI Agents SDK Multilingual DPI Voice Agent with Cartesia TTS

Non-English speaking citizens face accessibility barriers when navigating digital public services like Aadhaar and PM-Kisan portal updates. Build an interactive voice assistance agent using the OpenAI Agents SDK and Cartesia streaming TTS to provide low-latency audio support with under 500ms response latency.

Business OperationsHosted 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 a real-time voice assistance agent leveraging OpenAI Agents SDK function calling and Cartesia audio synthesis for regional language assistance.

Delivery guide

How work is evaluated

Evaluation

Evaluates response accuracy and audio response generation latency.

Datasets

Challenge datasets

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

Evaluation rubric

How submissions are scored

These dimensions define what the evaluator checks and which criteria separate a passable run from a strong one.

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

latency_check

Verifies latency is below 500ms threshold

Binary check

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

Dimension 2

latency_ms

End-to-end response generation latency in milliseconds • target: 450 • range: 0-2000

Binary check

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

Learning goals

What you should walk away with

  • Implement OpenAI Agents SDK runner loops with streaming audio integration

  • Connect Cartesia low-latency TTS for ultra-fast text-to-speech rendering in Hindi and Tamil

  • Build dynamic schema tools to query citizen service status endpoints

  • Ensure voice latency stays under 500ms end-to-end

Resources and assets

Reference links and supporting material

Dataset notes

Synthetic citizen service dataset containing portal update histories and localized voice queries.

Challenge setup

Evaluates response accuracy and audio response generation latency.

Challenge input

JSON audio transcript and target language

OpenAI Agents SDK

Agent control loops and function execution

Cartesia

Ultra low latency streaming voice engine

Evaluated output

JSON object with status, answer, and latency ms

Evaluation checks
  • Verifies latency is below 500ms threshold
  • End-to-end response generation latency in milliseconds • target: 450 • range: 0-2000
Configured evaluation evidence
  • Benchmark: Multilingual Voice Agent Benchmark
  • Latency Ms target: 450
  • 1 public reference case
Configured runtime
  • Python execution harness
  • Python sandbox (unavailable on Versalist)

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

Action Space
  • Cartesia · Required
  • OpenAI · Optional
  • Alibi Detect · Optional

Evaluation contract

  • latency_check · Weight 1
  • latency_ms · Weight 1

Metrics

  • latency_ms · Target 450

Benchmarks

  • Multilingual Voice Agent Benchmark
  • general

Reference cases

  • 1 public reference case

Workflow tasks

  1. Task 1

    Input: JSON audio transcript and target language

    Output: JSON object with status, answer, and latency ms

    Execution harness: Python

Evaluation tools

  • OpenAI Agents SDK · Agent Framework
  • Cartesia · Speech Synthesis

Recipe freeze marker

No recipe freeze time is recorded.

Frequently Asked Questions about Build OpenAI Agents SDK Multilingual DPI Voice Agent with Cartesia TTS