Skip to main content
Back to challenges
Challenge

Zoning Acceleration Agent with LlamaIndex and Cartesia Voice Interfaces

With Seattle eliminating environmental appeals to accelerate zoning changes and Singapore constantly issuing new site tenders, urban planners are overwhelmed with regulatory updates. This challenge tasks you with building a 'Zoning Acceleration Agent' using LlamaIndex. You will create a high-performance RAG (Retrieval-Augmented Generation) pipeline that ingests municipal PDFs, environmental appeal documents, and URA (Urban Redevelopment Authority) tender records. To make this tool accessible to field inspectors and site managers, you will integrate Cartesia to create a low-latency voice interface. The agent should be able to answer complex natural language queries about site-specific setbacks, floor area ratios, and recent environmental exemptions. Use LlamaIndex's advanced query engines to route questions between different policy documents and Cartesia's Sonic model to deliver real-time, low-latency audio responses for site queries.

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 specialized RAG agent for urban planning using LlamaIndex and implement a real-time voice interface via Cartesia for on-site zoning queries.

Delivery guide

How work is evaluated

Evaluation

Evaluates the RAG pipeline's accuracy in retrieving site-specific zoning data and the latency/quality of the voice output.

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

Citation Correctness

The agent must cite the correct document for each specific jurisdiction.

Binary check

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

Dimension 2

Retrieval Precision

Relevance of the top 3 retrieved chunks to the urban planning query. • 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

What you should walk away with

  • Implement advanced RAG using LlamaIndex for complex municipal and regulatory documents

  • Design a semantic routing system to distinguish between Singapore URA and Seattle zoning data

  • Integrate Cartesia's high-speed text-to-speech API for real-time assistant responses

  • Optimize data chunking strategies for legal and zoning documents to preserve context

Resources and assets

Reference links and supporting material

Dataset notes

A collection of 25+ PDF documents including Seattle's recent legislative changes to the State Environmental Policy Act (SEPA) and Singapore's URA site tender notices.

Challenge setup

Evaluates the RAG pipeline's accuracy in retrieving site-specific zoning data and the latency/quality of the voice output.

Challenge input

Natural language query text

LlamaIndex

Expertly handles complex PDF structures and multi-document retrieval.

Cartesia

Provides the low-latency audio needed for on-site field assistants.

Evaluated output

JSON with answer, source document citations, and TTS latency

Evaluation checks
  • The agent must cite the correct document for each specific jurisdiction.
  • Relevance of the top 3 retrieved chunks to the urban planning query. • target: 0.85 • range: 0-1
Configured evaluation evidence
  • Benchmark: Planning-RAG-Bench
  • Retrieval Precision target: 0.85
  • 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
  • Llama Index · Optional
  • Exa · Optional

Evaluation contract

  • Citation Correctness · Weight 1
  • Retrieval Precision · Weight 1

Metrics

  • Retrieval Precision · Target 0.85

Benchmarks

  • Planning-RAG-Bench
  • general

Reference cases

  • 1 public reference case

Workflow tasks

  1. Task 1

    Input: Natural language query text

    Output: JSON with answer, source document citations, and TTS latency

    Execution harness: Python

Evaluation tools

  • LlamaIndex · Data Framework
  • Cartesia · Voice Synthesis

Recipe freeze marker

No recipe freeze time is recorded.

Frequently Asked Questions about Zoning Acceleration Agent with LlamaIndex and Cartesia Voice Interfaces