Develop Voice-Enabled E-Commerce Delivery Support with Pydantic AI and Hume AI
Last-mile grocery delivery drivers in Indian metros face high order drop-off due to language constraints and ambiguous delivery addresses. Build a type-safe voice agent using Pydantic AI and Hume AI to parse spoken multilingual delivery updates and produce validated address corrections. Achieve a 92% extraction precision on voice transcripts.
What you are building
The core problem, expected build, and operating context for this challenge.
Develop an empathetic, multilingual voice interface using Pydantic AI and Hume AI to process rider voice interactions into structured logistics data.
How work is evaluated
Tests Pydantic AI schema outputs against acoustic transcripts and sentiment annotations.
Challenge datasets
Public datasets, plus the private datasets you have access to.
How submissions are scored
These dimensions define what the evaluator checks and which criteria separate a passable run from a strong one.
pydantic_validation
Validates output adheres to DeliveryUpdate model rules
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
extraction_precision
Precision of extracted address attributes • target: 0.92 • range: 0-1
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
What you should walk away with
Utilize Pydantic AI to build structured voice dialogue workflows
Integrate Hume AI expressive audio models for customer mood and delivery sentiment detection
Enforce strict schema validation on localized speech outputs in Hinglish and Tamil
Implement real-time delivery agent status updates into central routing systems
Reference links and supporting material
Dataset of driver audio transcripts and corresponding Hume AI vocal tone vectors recorded during urban delivery runs.
Challenge setup
Tests Pydantic AI schema outputs against acoustic transcripts and sentiment annotations.
Challenge input
JSON object with audio transcript and emotion vector
Pydantic AI
Ensures type safety and schema validation for model inputs and outputs.
Hume AI
Provides empathetic tone analysis and audio expression metrics.
Evaluated output
JSON object conforming to Pydantic DeliveryUpdate schema
- Validates output adheres to DeliveryUpdate model rules
- Precision of extracted address attributes • target: 0.92 • range: 0-1
- Benchmark: Multilingual Delivery Voice Extraction Benchmark
- Extraction Precision target: 0.92
- 1 public reference case
- 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
- Hume AI · Required
- Pydantic AI · Optional
- Libretto · Optional
- Pydantic AI · Optional
Evaluation contract
- pydantic_validation · Weight 1
- extraction_precision · Weight 1
Metrics
- extraction_precision · Target 0.92
Benchmarks
- Multilingual Delivery Voice Extraction Benchmark
- general
Reference cases
- 1 public reference case
Workflow tasks
- Task 1
Input: JSON object with audio transcript and emotion vector
Output: JSON object conforming to Pydantic DeliveryUpdate schema
Execution harness: Python
Evaluation tools
- Pydantic AI · Agent Framework
- Hume AI · Multimodal Model
Recipe freeze marker
No recipe freeze time is recorded.