Indeed is the world’s leading job site, where 27 people find jobs every minute. Behind those interactions is a powerful supergraph built on Apollo: more than 350 federated subgraphs handling over 60 billion operations monthly, with 200+ engineers shipping updates around 70 times a week.
When the company set out to build Talent Scout, a new AI-powered virtual hiring assistant, it had a bold goal: launch in nine months with zero prior AI experience. What made that possible was Indeed’s existing Apollo GraphQL infrastructure, which gave the team a ready-made foundation for innovation.
The Challenge: Build an AI Hiring Assistant Fast
Indeed’s employer experience team was tasked with creating a conversational AI that could help recruiters make better hiring decisions. The assistant needed to surface job performance metrics like impressions, clicks, and application rates. It had to suggest competitive salary ranges using Indeed’s market data and recommend improvements to job descriptions to attract stronger talent.
The timeline was aggressive: nine months from concept to launch. “Our team had no experience with AI. It was a learning experience to understand AI systems and build all the safeguards,” says Jeffrey Russom, Principal Software Engineer, who led the API design for the project.
With sensitive candidate data involved, every system powering Talent Scout had to be secure, auditable, and compliant. The team needed to focus their limited time on AI-specific challenges like preventing hallucinations and building safety guardrails, not on rebuilding data infrastructure.
Early on, the team realized their existing real-time system would not be enough. It was not secure, it was not type-safe, and it created a slow, clunky chat experience, not the kind of experience you want when helping employers make hiring decisions. They needed a new approach that could deliver structured, real-time updates reliably and safely.
The Advantage: What Apollo Made Possible
Without Apollo already in place, Indeed would have faced months of additional infrastructure work before even touching the AI components. The team would have needed to build streaming infrastructure from scratch, create a new orchestration layer for candidate and job data, and develop a backend-for-frontend API just for Talent Scout. They also would have had to integrate across multiple service protocols such as REST and gRPC.
“We’re talking about a lot of complexity in terms of how to integrate with potentially a myriad of different technologies just to get job and candidate data,” Russom explains. He estimates this infrastructure work would have added “several months, if not more” to the timeline.
Instead, the team had something better: years of GraphQL APIs already powering Indeed’s platform. Because those APIs already included permission checks and authorization logic, the team could trust that the same data governance applied automatically to Talent Scout’s AI queries. That allowed them to move quickly without compromising on safety or compliance.
They also inherited the visibility that comes with operating a large federated platform. With hundreds of services contributing to a unified graph, teams rely on predictable query behavior and clear ownership boundaries. Reusing this structure for Talent Scout gave the team immediate insight into data flow and eliminated the need to build separate monitoring or debugging workflows.
The Solution: Accelerating AI with Apollo
Indeed’s GraphQL-first strategy turned out to be exactly what the team needed. The team leveraged more than 40 existing subgraphs across Indeed’s Apollo supergraph. These subgraphs contained years of business logic around candidate data, job performance, and employer insights. The AI could tap into all of it without requiring any of those teams to rebuild their services.
“We were able to use the GraphQL APIs we had already invested in and spend our time learning how to use and govern AI, rather than worrying about how to make our data work with it” – Jeffrey Russom, Principal Software Engineer, Indeed
This separation of concerns was critical. While the team wrestled with AI-specific challenges like prompt engineering and hallucination prevention, GraphQL and Apollo Router handled all the complexity of data orchestration behind the scenes. The same permission framework that protects candidate data on Indeed’s platform applied automatically to the AI assistant, ensuring that the right data was served securely, every time.
“That’s where our partnership with Apollo really paid off. Apollo Router and Federated Subscriptions became the backbone of Talent Scout. Instead of polling or stitching together workarounds, we can now stream structured, real-time updates directly into the UI. We send progress messages, candidate insights, and multi-agent responses, all delivered securely through a federated graph.” – Jeffrey Russom, Principal Software Engineer, Indeed
The Results: From Concept to Launch in Nine Months
In September 2025, Talent Scout officially launched, onboarding 1,000 beta customers after a limited alpha with key partners. The product delivered a fully functional AI recruiter that helps employers evaluate candidates, analyze job performance, and optimize postings through natural conversation.
The impact was immediate. Conversations that once felt sluggish now feel instant. Employers see richer candidate information in real time, creating a more natural and interactive experience.
Key outcomes:
- 22% faster message response times
- 2 seconds faster delivery for multi-agent workflows
- One less service dependency, consolidated on Apollo Router
When the team needed real-time responsiveness, they adopted Apollo’s federated subscriptions, which allowed them to stream rich, structured data from existing subgraphs without requiring any other teams to change their APIs.
“Having Apollo and GraphQL allowed us to accelerate our AI work by providing that base capability of the data we needed for the AI agents,” says Russom. “We were able to free up our resources to focus on just the AI-specific problems.”
The Bigger Picture: A Foundation for Every Surface
For Russom, Apollo’s value extends far beyond accelerating one AI project. It solves a fundamental problem that has plagued the industry for years: how to consistently serve data across different client surfaces.
“Even if you’re a startup, if you want to target more than one client surface area such as web, native, or AI, you really should think about Apollo GraphQL. It lets you build a data catalog for all your clients and organize it in a way that makes sense across your organization.” – Jeffrey Russom, Principal Software Engineer, Indeed
The same subgraphs that power Indeed’s web and mobile applications now also support new AI agents, giving every surface a consistent and reliable source of truth. Because governance and permissions are built into the graph itself, the same policies apply whether the client is a recruiter using a web form or an AI agent generating insights.
Looking Ahead: Scaling AI with Apollo MCP Server
Indeed’s team is now exploring Apollo MCP Server, which helps AI agents access only the data they need from the graph.
“We’re just getting started,” Russom adds. “We’re already expanding this foundation to power collaborative features like shared candidate shortlists and document editing.”
Future plans include expanding Talent Scout to additional surfaces such as Slack and ChatGPT integrations, extending the same data foundation across new AI-driven experiences.
The Foundation That Keeps Paying Off
Indeed’s experience with Talent Scout shows how GraphQL infrastructure becomes more valuable over time. What began as an investment in API consistency became the backbone for a new generation of AI applications.
“Apollo is one of those investments that really started to pay off once we went GraphQL-first,” Russom reflects. “It solves real problems, and it’s what made Talent Scout possible.”
For teams looking to move quickly in AI without rebuilding their data infrastructure, Indeed’s story offers a clear takeaway: the right API layer doesn’t just connect systems. It accelerates what’s possible to build on top of them.
Want to go deeper? Jeffrey Russom shared the full technical story behind Talent Scout at GraphQL Summit 2025, and sat down for a detailed Customer Spotlight interview. Both sessions dive into the architectural decisions, real-world trade-offs, and lessons learned from building an AI assistant in nine months with zero prior AI experience.
Next story
Share article
