Perplexity-Powered Growth Intelligence Agent with OpenAI Agents SDK
Growth engineers, product marketers, and developer advocates constantly seek novel strategies to drive user acquisition, activation, and retention. This challenge tasks you with building an advanced autonomous agent using the **OpenAI Agents SDK** that leverages the **Perplexity Search SDK** as its primary research tool. The agent will act as a "Growth Intelligence Analyst," orchestrating sophisticated search-as-code queries to identify market trends, competitor strategies, ICP pain points, and nascent growth opportunities. Your agent must go beyond simple search. It needs to dynamically generate fan-out search queries, programmatically filter, deduplicate, and rank search results using custom Python logic, and score potential growth opportunities based on predefined criteria. The ultimate output will be a prioritized backlog of growth experiments, complete with actionable insights, suggested UTM parameters for tracking, and full citations linking back to the original search results. This requires deep understanding of OpenAI Agent tool integration, multi-step reasoning, and robust data processing within an agentic loop.
What you are building
The core problem, expected build, and operating context for this challenge.
Build an advanced OpenAI Agent that uses Perplexity Search to identify, score, and generate a prioritized backlog of growth experiments with UTM parameters and citations.
How work is evaluated
The solution will be evaluated based on the agent's ability to autonomously perform complex research, process information, and generate highly actionable, well-cited growth experiment suggestions. Key aspects include the agent's reasoning chain, the quality and relevance of search queries, the effectiveness of custom processing logic, and the structured completeness of its final output.
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.
Perplexity Tool Integration
Agent successfully calls the `perplexity_search` tool at least once.
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
Structured Output Format
The final output is a valid JSON object adhering to the `growth_experiment_backlog` schema.
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
Minimum Opportunities & Experiments
At least 3 distinct growth opportunities identified and 3 corresponding experiments generated.
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
UTM Parameters Present
Each experiment includes `suggested_utm_parameters` with at least `campaign` and `source`.
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
Precision@3 for Opportunity Relevance
Human evaluation of the top 3 identified opportunities for relevance to the initial prompt and target audience. • target: 0.8 • range: 0-1
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
Citation Completeness
Percentage of identified opportunities that have at least one valid, relevant citation (URL and snippet). • target: 0.9 • range: 0-1
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
Quality of Experiment Hypotheses
Human evaluation of the clarity, testability, and actionability of generated experiment hypotheses. Scored 1-5. • target: 4 • range: 1-5
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
What you should walk away with
Integrate the Perplexity Search SDK into an agent workflow that accepts an ICP, product focus, and competitor set.
Implement code-owned fan-out query generation, result filtering, URL/content deduplication, relevance ranking, and source citation preservation.
Convert ranked evidence into scored growth opportunities and a prioritized experiment backlog with UTM-ready campaign links.
Evaluate retrieval quality and output trustworthiness with precision-at-k, citation completeness, schema validation, caching, and rate-limit failure tests.
Reference links and supporting material
Sample data for 3 tasks
Challenge setup
The solution will be evaluated based on the agent's ability to autonomously perform complex research, process information, and generate highly actionable, well-cited growth experiment suggestions. Key aspects include...
Challenge input
```json { "initial_goal": "Identify 3-5 high-potential growth opportunities for a new 'AI-powered expense management SaaS' targeting small business...
OpenAI
OpenAI AI model provider
Zed
High-performance code editor
Coplay AI
AI coding assistant
Evaluated output
Agent's internal log/trace of tool calls, thought processes, and intermediate results.
- Agent successfully calls the `perplexity_search` tool at least once.
- The final output is a valid JSON object adhering to the `growth_experiment_backlog` schema.
- At least 3 distinct growth opportunities identified and 3 corresponding experiments generated.
- Precision@3 For Opportunity Relevance target: 0.8
- 3 public reference cases
- 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
- OpenAI · Required
- Zed · Optional
- Coplay AI · Optional
Evaluation contract
- Perplexity Tool Integration · Weight 1
- Structured Output Format · Weight 1
- Minimum Opportunities & Experiments · Weight 1
- UTM Parameters Present · Weight 1
- Precision@3 for Opportunity Relevance · Weight 1
- Citation Completeness · Weight 1
- Quality of Experiment Hypotheses · Weight 1
Metrics
- Precision@3 for Opportunity Relevance · Target 0.8
- Citation Completeness · Target 0.9
- Quality of Experiment Hypotheses · Target 4
Reference cases
- 3 public reference cases
Workflow tasks
- Task 1
Input: ```json { "initial_goal": "Identify 3-5 high-potential growth opportunities for a new 'AI-powered expense management SaaS' targeting small businesses, focusing on features or integrations that competitors lack.", "product_context": { "name": "ExpenseMind AI", "type": "AI-powered expense management SaaS", "target_audience": "Small businesses, startups", "current_features": ["receipt scanning", "categorization", "budget tracking", "integration with Xero/QuickBooks"] } } ```
Output: Agent's internal log/trace of tool calls, thought processes, and intermediate results.
Execution harness: Python
- Task 2
Input: Agent's final generated opportunities list.
Output: ```json [ { "opportunity_id": "OPP001", "title": "Integration with E-commerce Platforms for Expense Tracking", "description": "Opportunity to integrate 'ExpenseMind AI' with popular e-commerce platforms (e.g., Shopify, Etsy, WooCommerce) to automatically track business-related expenses incurred from platform sales/fees, a gap identified in competitor offerings.", "relevance_score": 9.2, "feasibility_score": 7.5, "impact_score": 8.0, "overall_score": 8.23, "citations": [{"source_url": "https://example.com/competitor-analysis-blog", "snippet": "...lack direct integration with Shopify for expense automation..."}, {"source_url": "https://example.com/ecommerce-trends-report", "snippet": "...small businesses increasingly rely on multiple platforms..."}] } ] ```
Execution harness: Python
- Task 3
Input: Agent's final generated experiment backlog.
Output: ```json { "growth_experiment_backlog": [ { "experiment_id": "EXP001", "related_opportunity_id": "OPP001", "title": "Pilot Program: Shopify Integration for ExpenseMind AI", "hypothesis": "Integrating ExpenseMind AI with Shopify will attract 15% more e-commerce small businesses within 3 months.", "description": "Develop a beta integration for Shopify users to automate expense tracking from their store activities. Recruit 20 pilot users for feedback.", "metrics_to_track": ["new sign-ups from e-commerce", "integration usage rate", "user satisfaction (NPS)"], "suggested_utm_parameters": { "campaign": "shopify_integration_pilot", "source": "blog_post", "medium": "email", "content": "early_access_signup" } } ] } ```
Execution harness: Python
Recipe freeze marker
No recipe freeze time is recorded.